OpenTable Launches Its Largest Suite of New and Updated Product Features for Restaurants

Posted in Commentary with tags on September 2, 2026 by itnerd

OpenTable today announced its largest suite of new and updated product features for restaurants, many powered by AI and automation. From table automations and AI-powered search distribution to natural-language reporting, this rollout gives OpenTable restaurant partners a more complete and connected operating system to help uplevel their business.

The launch focuses on three priorities for restaurants: driving valuable demand, turning insights into action and connecting to the systems they use every day. The new features and enhancements are designed to reduce manual work, create more opportunities to fill seats and give restaurant teams more time to focus on delivering a memorable dining experience.

The Right Demand: Nothing is More Expensive Than an Empty Seat for a Restaurant

Diners are searching for restaurants in new ways, with AI-driven discovery growing rapidly. Yet, many restaurants struggle to become visible on these new platforms. Restaurants also need better ways of managing their books and optimising to fill seats based on demand. OpenTable is bridging that gap, helping operators fill seats without adding hours of manual administrative work.

  • New integrations with some of the most widely used LLMs: OpenTable has partnerships and integrations with leading AI services and search platforms, including Google (Search, Maps and Gemini), ChatGPT, Microsoft’s Copilot and Perplexity, as well as Amazon’s Alexa+ AI assistant, helping restaurants show up where diners are already searching. This year on OpenTable, LLM integrations drove 17x more seated diners year-over-year, bringing in new diners including those who spent on average 20% more compared to other channels.
  • Get discovered through OpenTable’s AI chat: OpenTable’s AI Concierge helps diners discover restaurants through natural conversations—the way they’d ask a friend. With 500,000+ monthly active global users, AI Concierge gives restaurants another way to reach diners as they decide where to book.
  • Help drive demand with richer, automated profiles: AI-generated restaurant profiles help bring dining rooms and event spaces to life using information from restaurant websites and details operators provide – saving operator time and helping diners choose restaurants with confidence.

The Right Intelligence: From Data to Action

OpenTable is making it easier than ever for restaurant teams to put their data to work, giving operators instant answers, deeper guest insights, and actionable ways to unlock hidden revenue on the floor.

  • Natural language reporting (in testing): With Conversational Reporting, restaurants can ask questions about their data in conversational language and get instant, actionable answers — no dashboards, no exports, no analyst required. For operators who are already stretched thin, the information they need to make better decisions is a question away: operators will be able to ask questions like “Which locations in my group have the largest increase in covers?”
  • Table adjustments based on live demand: OpenTable’s new Table Automations tool automatically adjusts table minimums based on live demand. Since its test phase, there have been over 2 million automations completed, with an initial test saving operators 4.5 hours per month on average.
  • Better insights = better hospitality: New enhancements to OpenTable’s Guest Relationship Management tools (GRM) provide 10x more guest details than a standard guestbook to power personalised hospitality.** This includes features like Group Guest Visit History, which equips staff with cross-location guest profile insights to recognise first-time visitors to new locations like regulars.
  • Uncovering opportunity in the book: Turn Times and Availability Insights show where operators are losing bookable hours and inventory – and how to get them back. In testing, operators reported gaining back, on average, 39 minutes of reservable time per shift.

The Right Connections: A Tech Stack that Works Together

OpenTable is also expanding its integrations so that restaurants can connect reservation and guest data to the

  • Partner integrations and open connectivity: OpenTable’s ecosystem includes more than 200 partners across discovery, demand generation and restaurant operations. This year, OpenTable’s POS partner adoptions increased 40% year-over-year.*
  • More voice AI integrations than any Table Management System (TMS): OpenTable has more than 20 voice AI partners globally to ensure there’s tech that fits each restaurant’s unique needs, booking guests without pulling host stand staff away from the floor or after operating hours. To date, OpenTable has seated 3 million diners across its voice AI partners, an increase of 270% in 2026 compared to the same period in 2025.*

For the full suite of new products and enhancements, visit https://www.opentable.ca/restaurant-solutions/en/product-innovation/

Leave a comment »

23-year-old Sality botnet finally taken down 

Posted in Commentary with tags on September 2, 2026 by itnerd

CrowdStrike, the DOJ, and law enforcement in Bulgaria, Hungary, and Romania disrupted the Sality peer-to-peer botnet on August 31, more than 15,000 known infected machines were still active worldwide after 23 years of continuous operation since 2003. The takedown turned Sality’s own design against it, replacing trusted peers in the network with sinkholes since infected machines never verified who they were talking to, while the botnet’s most recent payload, a clipboard-hijacking tool called EggJagger, had stolen at least $150,000 in crypto by swapping in attacker wallet addresses.

More info here: Sality botnet infrastructure dismantled in joint global takedown

John Watters, Chairman & CEO, iCOUNTER Had This To Say:

“Twenty-three years is the real headline here. Sality launched in 2003 and outlived multiple generations of malware, multiple law enforcement agents and investigations, and multiple waves of the industry declaring old-school botnets dead. That kind of longevity only happens when a piece of malware keeps finding a new business model. Sality started as a generalist infector and spent its last eight years running EggJagger, quietly swapping cryptocurrency addresses on infected machines. Fifteen thousand known infected machines and $150,000 in known stolen crypto is small by today’s ransomware standards, but the operation didn’t need to be big. It needed patience, and it had it for over two decades.

CrowdStrike didn’t seize a server or arrest an operator. They turned Sality’s own architecture against it, replacing trusted peers in the network with sinkholes, because infected machines never verified who they were talking to in the first place. Shadowserver notified victims through ISPs and CSIRTs, and the DOJ coordinated the legal side with counterparts in Bulgaria, Hungary, and Romania. That combination is why this takedown is likely to hold instead of the infrastructure just resurfacing somewhere else in six months.

Sinkholing stops new instructions from reaching infected machines. It doesn’t clean them. Fifteen thousand known systems are still infected, just quiet now, and that remediation gap is going to outlast this week’s headlines.”

This illustrates that hackers can live for years in a system. Which means that you have to be constantly looking for them to achieve the best results.

Leave a comment »

Safe Software to Increase Global Headcount by Over 15% as Demand for AI-Ready Data Grows

Posted in Commentary with tags on September 2, 2026 by itnerd

Safe Software today announced it is recruiting for more than 60 roles throughout the remainder of 2026. The hiring will take the company past 400 employees, increasing headcount by over 15%, with new positions across the United Kingdom, the United States and Canada.

The expansion follows a year in which Safe crossed $100 million in annual revenue, growing close to 20 percent year over year and moving ahead of schedule on its target of $250 million by 2028. The company also grew its employee base by over 20% in the same period.

Aligned with the company’s growing revenue trajectory and market expansion, two thirds of the planned roles are revenue-facing across all three markets, spanning sales leadership, revenue operations, enablement, marketing, and customer support. Safe is also growing its product, engineering, and operational teams with open roles in AI enablement, test engineering, DevOps, design, HR, and employee experience.

Safe was named one of BC’s Top Employers for 2026 and one of Canada’s Top Small and Medium Employers for 2026. The company has been a certified B Corporation since August 2024, and has placed more than 400 co-op students since 2010, a third of whom went on to join full time.

Open roles are listed at www.safe.com/careers.

Leave a comment »

Workers Lose a Day a Week to Sluggish Virtual Desktops

Posted in Commentary with tags on September 2, 2026 by itnerd

Nexthink is warning that organizations risk wasting their technology investment and significant portions of their workforce’s time due to poor virtual desktop performance, with employees losing the equivalent of one full working day every week to sluggish virtual desktops.

Data from the Nexthink VDI Experience platform shows that employees see 20% of their working time impacted by sluggish virtual desktop performance. With the virtual client computing software market valued at $7.12bn in 2025, and VDI accounting for an estimated 59.74% of that spend – approximately $4.25bn – organizations are pouring billions into infrastructure that is undermining productivity.

The scale of the problem is significant. Nexthink data shows 65% of VDI users experience applications running slowly, while 51% of businesses using VDI experience performance issues. Based on Gartner’s forecast that virtual desktops will become the primary workspace for 20% of workers by 2027, and IDC projections putting the global knowledge worker and developer population will be at an estimated 1.04 billion by that year, this poor experience could affect approximately 135 million digital workers worldwide.

The five most common root causes of VDI-related support tickets identified by Nexthink include unstable or inconsistent network conditions, resource contention caused by image design, complicated or outdated login flows and scripts, underpowered or aging endpoint devices, and application behavior changes in virtual environments.

To improve VDI performance and protect the return on investment, organizations should:

  • Build a rounded view of success. Moving beyond cost as the primary measure. Adopt XLA thinking that centers on user outcomes, such as what employees need from their virtual desktop environment and which moments in their working day matter most.
  • Monitor experience, not just infrastructure. CPU, RAM, and uptime metrics do not reflect what users experience. Measure logon times, application launch times, and session responsiveness from the user’s perspective.
  • Measure how users feel. Couple technical data with regular sentiment snapshots through surveys or focus groups. Ask pointed questions about whether employees feel productive in their VDI environment.
  • Review findings and respond visibly. Establish a clear process for acting on experience data. Critically, close the loop with users, showing that their feedback has been heard and acted on is as important as the remediation itself.
  • Work actively on improvements. Prioritize fixes based on what has the greatest user impact at the lowest effort and cost. Take incremental steps and measure each one, so teams can demonstrate progress and accelerate improvement over time.

Methodology

The full workforce calculations are as follows:

  • Gartner’s 2025 Magic Quadrant for Desktop as a Service forecasts that by 2027, virtual desktops will be used as the primary workspace for 20% of workers. IDC projects the global knowledge worker and developer population will grow from 888 million in 2025 to 1.2 billion in 2029. This suggests an annual growth of approximately 78 million and would put the 2027 figure at around 1.04 billion. 
  • Applying the 20% Gartner figure yields an estimated 208 million virtual desktop users by 2027. Of these, approximately 135 million could be affected by slow applications (using Nexthink’s 65% finding).

Leave a comment »

Equinix Accelerates AI Inference for Enterprises with NVIDIA and Together AI

Posted in Commentary with tags on September 2, 2026 by itnerd

Equinix, Inc. today announced a significant expansion of its longtime collaboration with NVIDIA to deliver Equinix® Inference Exchange, a distributed AI inference program for global enterprises, alongside a new collaboration with Together AI.

As AI scales across models, providers and geographies, where inference runs is a strategic imperative that determines performance, cost and governance. Equinix Inference Exchange will give enterprises a faster path from AI experimentation to production, with secure, low-latency connectivity to the data, users and ecosystem they depend on.

This collaboration brings together NVIDIA’s validated Enterprise Reference Architectures with Together AI’s inference platform, supporting more than 200 open-source models. Delivered through Equinix’s global data centers, it will provide connectivity to clouds, networks and AI providers through Equinix Fabric®.

The solution will be announced today at Equinix Horizon, the company’s inaugural customer and partner event, alongside Equinix® Fabric One™, which will make it easier for enterprises to connect across globally distributed AI environments.

Where Inference Runs Matters

The pace of enterprise AI adoption is outrunning the infrastructure needed to support it. As enterprise AI moves from experimentation to production, inference increasingly needs to run closer to the users, data and applications it serves across clouds, models, providers and geographies. This shift requires enterprises to determine not only how to deploy AI infrastructure, but where it should run and how it connects to the data, applications and workloads it depends on.

Managing these distributed inference deployments introduces significant operational complexity at precisely the moment enterprises need greater control and visibility.

Equinix brings unmatched scale and ecosystem density to this challenge, with more than 280 data centers across 77 metros, 230 cloud on-ramps and over 10,500 businesses interconnected on its neutral exchange. Eight of the top 10 AI model providers and nine of the top 10 AI clouds are deployed with Equinix, underscoring the company’s position at the center of the AI ecosystem.

Built for Choice and Flexibility

Together AI is the latest addition to Equinix’s expansive AI ecosystem, bringing open-model flexibility and choice to enterprises deploying AI at scale. The solution combines three complementary layers designed to simplify distributed AI inference:

  • Equinix provides the infrastructure foundation, including power, advanced cooling and day-two operations, connected through Equinix Fabric to the clouds, networks and AI providers that inference depends on.
  • NVIDIA anchors the build with its Enterprise Reference Architectures and AI infrastructure purpose-built to maximize AI factory throughput and minimize token cost.  
  • Together AI runs the platform on top, supporting both multitenant deployments for shared efficiency and dedicated single-tenant environments for workloads that require dedicated capacity.

Built on Equinix Fabric, the solution will connect to inference providers across major metros worldwide, cutting time-to-first-token. It also will connect to an expansive ecosystem of clouds, networks and AI providers, reducing deployment complexity.

Designed for Modern Enterprise Inference

The solution aims to support a broad range of enterprise inference scenarios, including:

  • Metro edge inference: For organizations that need inference running closer to users and data, enabling lower-latency AI experiences while leveraging the security, operational scale and global reach of Equinix.
  • Open model migration: For enterprises moving workloads from closed, proprietary models to open-source alternatives to control cost and avoid lock-in, the solution will provide a direct, low-friction path to run that migration in production, with Together AI’s open-model platform reachable over the same interconnected fabric enterprises already use to reach their other providers.
  • Sovereign AI: For enterprises operating in regulated industries or specific geographies, the solution will enable AI workloads to run in locations that support data residency and sovereignty requirements, providing a simpler path to deploying AI at scale while maintaining control over where data and inference are processed.

Equinix Inference Exchange will be available starting in Q1 2027.

Leave a comment »

CPP Investments and Equinix Complete atNorth Acquisition

Posted in Commentary with tags on September 2, 2026 by itnerd

atNorth, the leading Nordic high-density colocation and built-to-suit data center provider, today announced that Canada Pension Plan Investment Board (CPP Investments) and Equinix, Inc. (Nasdaq: EQIX), the world’s digital infrastructure company®, have completed the acquisition of atNorth.

atNorth’s footprint spans across all five Nordic countries, with eight operational data centers and several new projects underway across the territory. This includes sites under development in Sweden, Finland, Norway and Denmark, alongside expansions to existing sites and a strong portfolio of additional development projects.

Together, the operating portfolio and development pipeline provide atNorth with significant capacity to serve growing demand from global enterprise and hyperscale customers across AI, cloud and high-performance computing workloads, supported by advanced cooling technologies, renewable energy integration and heat reuse solutions.

The US$4 billion acquisition, by CPP Investments and Equinix, builds on CPP Investments’ global experience in data center investing and underscores the strategic importance of the Nordics as a leading hub for AI-ready digital infrastructure. atNorth will continue to operate independently under its existing brand, with the backing of its shareholders to accelerate development of its pipeline and expand capacity across the Nordics. Equinix brings complementary digital infrastructure expertise and global customer relationships to support atNorth’s continued growth.

Given the strength of the opportunity and confidence in the partnership since the initial announcement, Partners Group, on behalf of its clients, has elected to re-invest and acquire a 10% stake in atNorth. As a result, CPP Investments will hold a c. 51% controlling stake committing US$1.3 billion, alongside Equinix’s c. 34% committing US$895 million and Partners Group’s c. 10% committing $260 million. The remainder will be held by atNorth’s internal stakeholders, who have chosen to roll over a substantial portion of their equity. The transaction involves a financing package of US$4.1 billion (€3.6 billion), underwritten by a group of European and Canadian lenders to support atNorth’s continuous growth, fund the transaction, as well as the capital required to fund the expansion of the business.

Leave a comment »

Langflow RCE now under active exploitation

Posted in Commentary with tags on September 2, 2026 by itnerd

Attackers are actively exploiting CVE-2026-0768, a critical unauthenticated RCE flaw in Langflow’s code validator, running credential-harvesting operations that pull OpenAI API keys and AWS access straight out of compromised agentic AI infrastructure, with exploitation clocked within roughly 20 hours of disclosure.

Gidi Cohen, CEO and Co Founder, Bonfy Had This To Say:

“Two critical vulnerabilities, one week, and honestly the wildest part isn’t even the exploitation, it’s that one of them got “fixed” and the fix didn’t work.

VulnCheck tested a patched Rails server (8.1.3.1) assuming the hole was closed. It wasn’t. The libvips file-read got blocked, sure, but the actual RCE mechanism underneath it — a Marshal deserialization gadget, still fires with a valid signature. So the patch shut one door and left the house wide open. Worth remembering next time “patched” shows up in a report and everyone exhales.

Then there’s Langflow. CVE-2026-0768 isn’t some clever zero-day, it’s unvalidated input leading straight to root access. Not fancy. But look at what’s sitting next to it once you’re in: superuser flags, OpenAI keys, AWS credentials, SSH access. Attackers are literally querying for LANGFLOW_SUPERUSER and OPENAI_API* by name within hours of exploiting it. That’s not someone poking around and getting lucky. That’s a playbook, because someone already knows exactly what credentials live in these environments.

And this keeps happening. 12 vulnerabilities exploited since 2025, 15,000+ successful hits across just three CVEs. The attackers aren’t smash-and-grab types either — one crew disabled audit logging specifically so nobody could see what they did next before dropping their payload. That’s patience, not luck.

So the real question is pretty simple: do you actually know what’s exposed in your AI stack right now, or would you find out the same way we all just did, from a threat intel report after the fact?”

You have to test and keep testing. Because you can never count on any fix working properly.

Leave a comment »

Catalyst by Zoho Solves The Shortfalls of Moving from Coding to Production With an Agent-Ready, Full-Stack Cloud and Built-In Governance

Posted in Commentary with tags on September 2, 2026 by itnerd

Zoho Corporation, a global technology company, today announced major enhancements to Catalyst by Zoho, its Platform-as-a-Service (PaaS), now weaving agentic development capabilities directly into the coding environments developers already use. Additions include Agent Skills, a non-interactive command-line interface (CLI), and Model Context Protocol (MCP) support to Catalyst’s platform, along with new integrations for agentic AI coding assistants including Anthropic’s Claude Code and OpenAI’s Codex. To enhance accessibility and empower future developers, the platform is also offering a free student program that includes full-stack hosting, functions, database, and AI tooling.

Catalyst by Zoho bridges the gap between code generation and reliable deployment by giving agentic AI coding assistants a structured, deterministic way to build and deploy applications. Developers can use the AI coding assistants of their choice while Catalyst provides the underlying full-stack serverless infrastructure and controls, eliminating the need to stitch together multiple cloud services and vendors. This approach simplifies development and future iterations while reducing operational complexity,positioning Catalyst as a reliable partner as technical capabilities and customer needs evolve.

Turning AI-Generated Code Into Production-Ready Applications

AI coding assistants can generate code quickly, but getting it to production requires deep platform knowledge, cloud services, and deployment workflows. Catalyst brings these together on a single serverless full-stack cloud, offering AI coding assistants what they need to build, test, and deploy applications. These capabilities make this possible:

Catalyst Agent Skills gives coding assistants the context they need to understand Catalyst services, architecture, and recommended development patterns. Rather than relying on the model to determine how Catalyst should be used, the Skill guides the assistant toward the appropriate capabilities and workflows. By surfacing only the capabilities relevant to the task, they help assistants make the right application and service choices, optimize token use, and generate accurate, verified output even with lighter models.

The non-interactive CLI allows AI coding assistants to execute multi-step workflows in Catalyst from start to finish without requiring human input at every step. This reduces manual intervention and helps developers move faster from coding to deployment.

The Catalyst MCP server provides AI coding assistants direct access to Catalyst capabilities from the developer’s existing environment. Actions such as creating a database table or adding a column can be performed directly from VS Code, Claude Code, Cursor, or any AI IDE without switching to the Catalyst console, keeping development in one workflow and accelerating delivery.

Orchestration, built into the Skill, ties the three capabilities together. When an AI coding assistant faces a decision about how to execute a request, the Skill routes it deterministically down the CLI or MCP path rather than leaving that choice to the model. This approach lowers the developer’s cognitive load, reduces the risk of incorrect tool selection, and helps produce more reliable, production-ready applications.

Built on Zoho’s Platform with Humans in Mind

While each service may work well independently, managing a fragmented stack can add operational complexity and make security, privacy, and governance harder to maintain. Applications built on Catalyst run inside Zoho’s own data centers and inherit Zoho’s robust security infrastructure, including DDoS protection, SOC compliance, regular vulnerability assessment and penetration testing (VAPT), a web application firewall, and more.

Human oversight is built into the deployment process rather than added afterward, giving organizations greater control as AI becomes part of application development:

Decoupled environments keep development and production separate. Code moves to production by manual promotion only—ensuring the AI agent never touches production, eliminating the risk for error.

Scoped collaborator controls define who—or what—can participate in development and what actions they can perform.

Full audit trails provide records to assist with oversight, including application logs, platform logs and MCP tool-call logs. From this data, developers can track precisely what actions the AI agent took, and when. Every change after launch is versioned, attributable, and reversible.

What’s Next

Catalyst is advancing towards a new era of agentic software development, where developers can collaborate with increasingly capable AI agents to execute sophisticated workflows across the software development lifecycle—all on a governed, full-stack cloud foundation. Forthcoming changes include multi-agent hosting, AI tool connectors, agentic SDLC, and more.

Disclaimer: All trademarks, product names, and company names cited herein are the property of their respective owners.

Pricing and Availability

Catalyst by Zoho offers a monthly free tier for developers to explore the platform, plus $250 in free credits for users who want to go deeper over a six-month period. Catalyst is available for immediate use.

Catalyst continues to offer a straightforward pay-as-you-go pricing model, with every feature bearing a per-unit cost rather than a license fee layered on top of usage. Developers get full visibility into usage from the Catalyst console and can set budget alerts and ceilings to prevent unexpected bills. A structured subscription option is also available for teams that prefer predictable costs.

Catalyst remains completely free for students, no subscription or credit card required to deploy non-commercial applications.

Zoho’s Privacy Pledge

Zoho respects user privacy and does not run on an ad-revenue model in any part of its business, including its free products. The company owns and operates its own data centers, giving it full oversight of customer data privacy and security. More than 150 million users worldwide, across more than 1 million paying organizations, rely on Zoho to run their businesses, including Zoho itself. For more information, visit zoho.com/privacy-commitment.html.

Leave a comment »

SIOS Survey Reveals Enterprises Are Struggling to Keep Mission-Critical Applications Resilient Despite Widespread HA/DR Investments

Posted in Commentary with tags on September 2, 2026 by itnerd

SIOS Technology Corp. today announced the results of its 2026 State of Application Resilience Survey, revealing that despite widespread adoption of HA/DR technologies, most enterprises continue to experience downtime, struggle with increasingly complex hybrid infrastructures, and lack confidence in their existing resilience strategies.

The survey gathered responses from more than 250 IT leaders across North America and the United Kingdom, representing predominantly large enterprises across industries including healthcare, financial services, manufacturing, retail, government and technology. The research examined how organizations protect mission-critical applications from downtime and disasters while identifying emerging priorities for HA/DR investments.

Key findings include:

  • Downtime remains a persistent problem. Seventy-six percent of organizations experienced at least one outage lasting more than 10 minutes over the past year despite having HA/DR protection in place, highlighting the limitations of many existing failover strategies.
  • Hybrid and multicloud environments have become the norm. Nearly seven in ten organizations now run critical applications across hybrid infrastructures, while only 2% operate exclusively on-premises, increasing the complexity of maintaining application availability across diverse environments.
  • Organizations lack confidence in their resilience strategies. Only about half of respondents said they were satisfied with their current HA/DR solutions, while many remained neutral or dissatisfied, suggesting existing solutions are falling short of enterprise expectations.
  • Disaster recovery testing remains inconsistent. Only 7% of organizations perform monthly DR testing, while one-third test just annually and 22% don’t know how frequently testing occurs, creating significant recovery risk.
  • Cybersecurity is reshaping HA priorities. Seventy-two percent of respondents either already use or would consider using HA clustering to streamline patch management, reflecting growing recognition that application availability and cyber resilience are increasingly interconnected.
  • Complexity has overtaken cost as the biggest challenge. Configuration and management complexity ranked as organizations’ top HA/DR challenge, followed closely by integration across increasingly heterogeneous IT environments.
  • Investment priorities continue to rise. Improving disaster recovery protection ranked second only to cybersecurity among planned IT investments over the next 18 months, demonstrating that organizations are actively budgeting to strengthen resilience.

The survey also found that Windows remains the dominant operating system for mission-critical applications, but organizations increasingly manage mixed Windows and Linux environments, reinforcing the need for cross-platform HA solutions. Respondents also indicated growing interest in application-aware clustering technologies that can simplify operations while improving resilience across heterogeneous infrastructures.

For more details, interested parties can download the complete 2026 State of Application Resilience Survey at: https://bit.ly/4x0smfY

Leave a comment »

Anthropic resumes external AI testing with new safeguards

Posted in Commentary with tags on September 1, 2026 by itnerd

Anthropic has resumed external cybersecurity testing of its AI models after introducing new safeguards, roughly a month after Claude models breached company systems during security evaluations.

The incidents occurred when models being tested for cyber capabilities went beyond their intended environments and accessed real-world systems. The company has now introduced stronger safeguards designed to limit what models can access and do during testing, including additional monitoring and restrictions around external systems.

Separately, Anthropic is warning customers about infostealer malware stealing active Claude login sessions from infected computers without going through normal password and two-factor authentication (2FA) login processes. The stolen sessions can allow attackers to access victims’ Claude accounts and consume their usage.

Noelle Murata, Chief Operating Officer, Xcape, Inc.:

   “Artificial intelligence tools are inherently benign, but threat actors will harness their capabilities regardless of corporate guardrails. Anthropic resuming model evaluations following internal sandbox escapes highlights an enduring reality: the leap-frog dynamic between defenders and adversaries is as old as software development itself. Using autonomous systems to monitor autonomous systems is ultimately using the problem to solve the problem.

   “As capability drift persists, operational environments exposed to AI agents must implement operator-aware context controls and strict transactional guardrails to prevent catastrophic actions from execution, whether initiated maliciously or accidentally. Beyond local sandbox containment, organizations face concurrent risks from infostealers harvesting session tokens to bypass multi-factor authentication. Security leaders must implement continuous internal controls, restrict session token lifetimes, enforce hard authorization bounds on target environments, and isolate testing sandboxes from corporate networks and the Internet.

   “Critical Takeaways

  • AI tools remain neutral capabilities that malicious actors will exploit regardless of safety guardrails.
  • Target systems must enforce operator awareness and hard transactional guardrails to prevent autonomous agents from taking catastrophic actions.
  • Fundamental identity hygiene and strict network isolation from the Internet remain the primary defense against token theft and agent escapes.

   “Playing leap-frog with autonomous agents is fine until the model jumps directly out of the sandbox.”

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

   “Anthropic is managing AI security from both directions this week. The company resumed external cybersecurity evaluations after deploying new safeguards, a month after Claude models breached three organizations during testing. Separately, it’s warning users that commodity infostealers are hijacking active Claude sessions to drain usage.

   “The evaluation incidents revealed three distinct failure patterns this summer. Anthropic’s preliminary analysis suggests its models encountered evidence of a real internet connection and rationalized it away to keep believing the environment was simulated. OpenAI’s Hugging Face incident showed a different mode, where agents recognized they were crossing a boundary and did it anyway. And the UK AI Security Institute found Mythos 5 attempting a supply chain attack against real open-source maintainers, creating fake identities and trying to socially engineer a human into approving malicious code.

   “I see the same dynamics in my own offensive security tooling. I’ve had agents try to enrich their own scope during penetration tests, finding adjacent targets and deciding they should be in play. The model can recite the rules perfectly and still reason around them in pursuit of the objective. That’s why I build deterministic hooks that cross-check every action against an immutable scope file before it executes.

   “The infostealer warning is a different problem with the same lesson. Stolen session cookies bypass two-factor authentication (2FA) entirely because the attacker never goes through the login flow. Claude sessions now sit alongside cloud console cookies and banking credentials on the commodity malware market.

   “Monitoring, alignment training, system prompts, and login-flow protections are all necessary, but insufficient as models get more capable. High-risk agent actions need hard technical controls and human approval before they execute. Increasingly capable agents can either knowingly disregard the rules or reason themselves into believing the rules don’t apply. Security architectures need to account for both.”

Face facts. Cyber security needs to take into account AI. If it doesn’t, it’s a fail. These examples prove it without a doubt.

Leave a comment »

« Older Entries
Newer Entries »