The IT Nerd

The Abstract ASTRO team issued this blog post re: PHP2Shell and the Critical PHP RCE Vulnerabilities CVE-2026-109618 and CVE-2026-109619. ASTRO’s post focuses on what is confirmed, what organizations can do before the patch lands, and the detections they can put in place. The post will be updated as more information becomes available and can be disclosed.

You can read the deails here: https://www.abstract.security/blog/php2shell-critical-php-rce-vulnerabilities-cve-2026-109618-and-cve-2026-109619

Abstract today announced new AI SOC capabilities in its agentic SIEM: investigation, hunt, and triage powered by a combination of Abstract’s own trained and tuned models and frontier foundation models, and SOAR response capabilities with actions assigned directly to detection rules. The capabilities are built into the platform that already collects, routes, and detects on security data in stream.

Security teams are facing agentic attacks that move at machine speed while their own tooling spreads the response across a pipeline vendor, a SIEM, a SOAR, a threat intelligence feed, and an AI triage bolt-on. Dozens of AI SOC vendors sell agents that sit on top of someone else’s alerts and add another contract to the stack. Abstract runs the entire loop on one platform: 400+ integrations feed streaming detection that operates in real time, before data lands, findings combine into entity-based insights, and the SIEM itself now investigates, classifies, and responds on the same data plane.

Most AI bolted onto the SOC re-enriches the same alert from scratch every time it looks at one. Abstract’s pipeline builds an enterprise context layer that makes the data AI ready before the alert ever fires: events are normalized and enriched in stream with identity, threat intelligence, and asset context. That normalization gives AI the security semantic layer it needs, so a field like src is never confused between the sender of the data, the host that opened the connection, or the internal host another source means by it, and every investigation starts from context the platform already trusts.

The agentic SIEM works every alert in the queue. Alerts that once aged out or got closed in bulk to keep the queue moving now each get an investigation with evidence attached, and analysts review conclusions instead of raw alerts.

The new capabilities span the full security operations loop:

• Investigation and triage run on a combination of Abstract’s own proprietary trained and tuned models and frontier foundation models. The ASTRO threat research team trains Abstract’s models, baking the team’s real-world intelligence into the artificial intelligence of the platform and keeping investigations current with the latest threats the team tracks in the field.
• Real-time classification. Abstract’s models return triage decisions in real time, keeping analysis moving at the pace of the stream and feeding response actions while activity is still underway.
• Threat hunting runs on the same data plane as detection and investigation, across the identity, cloud, endpoint, SaaS, and network data already flowing through the platform.
• SOAR capabilities. Automated response is built into the platform, with containment, enrichment, and notification actions firing the moment a detection matches.
• Investigations Astro runs today include threat intelligence detections, endpoint agent analysis, insider risk analysis, phishing and suspicious mail triage, and cloud attack analysis.

Where foundation models trained on the open internet bring broad knowledge and unvalidated confidence, Abstract’s models ground each assessment in the ASTRO team’s validated threat expertise, so verdicts reflect how attacks actually unfold rather than how they read in public write-ups. Foundation models handle the deep reasoning on cases that warrant it, with the full evidence trail preserved.

Availability

The new AI SOC capabilities are generally available today as part of the agentic SIEM. Investigation bundles are available as add-ons sized by alert volume. Existing customers can add investigation bundles through their Abstract account team. Details at https://abstract.security.

The Abstract ASTRO team have been studying AI attack agents with Eyal Sela & Gambit Security and have found some interesting playbooks. They just blogged about this late yesterday evening.

One small yet interesting piece: a method for finding the real server behind your CDN. They check SPF records, the staging subdomain someone forgot to proxy (oops), and your favicon hash sitting in Shodan. Once they find your origin, they go straight to it and your WAF never gets a vote.

They’re tidy, too. Tools get shredded, logs get scrubbed, and timestamps get backdated before you’ve finished your coffee.

But the nice thing about robots is they’re lazy in very predictable ways. And their cleanup routine reads like a checklist of commands your web server should never run. Thanks for the detection list, guys!

It’s all written up: a script to find your own origin leaks before they do, plus managed detections that fire while the attacker is still on the box (not three days later in a log review).

Homework for this week: does your origin only accept 80/443 from your CDN’s ranges? If you’re not sure, the answer is probably no.

As you probably have already seen, a security researcher known as Chaotic Eclipse has published a proof-of-concept exploit called FalconFlank targeting CrowdStrike’s Falcon endpoint platform — the latest in a pattern of public releases against major endpoint security products, following recent disclosures against Kaspersky Endpoint Security and Gen Digital’s Avast Antivirus.

Abstract’s ASTRO research team blogged about it this afternoon and here are some highlights:

• The exploit turns Falcon’s own defenses against it. FalconFlank reportedly abuses Falcon’s Microsoft Office malicious-macro-removal feature. It’s a remediation function that runs with elevated privileges and escalates from a low-privileged local user to a more powerful context on fully patched Windows 11 25H2 and Windows Server 2025 systems.
• No CVE, no vendor confirmation yet. As of the blog’s publication, CrowdStrike has not confirmed the flaw’s validity or scope, and there’s no assigned CVE. The technical claims are researcher-provided and currently unverified.
• A track record worth noting. This researcher has previously released working exploits, largely against Microsoft products including Windows and Defender, with some later observed being abused in the wild…often following public criticism of vendor vulnerability-handling processes.
• Defenders aren’t stuck waiting on a patch. Abstract ASTRO has published behavioral detection logic, keyed on the underlying mechanism (an anomalous OLE file write, a specific DLL artifact, and a directory-mirroring/coordination pattern) rather than the exact published sample, which means it should hold up even against evasion attempts the researcher has already flagged.

Abstract points out that it’s tricky wth EDR products: the privileges these agents need to protect a host are the same privileges that create new attack surface when abused.

UPDATE: A CrowdStrike spokesperson said:

“We are actively investigating these claims and advise customers to disable the Microsoft Office File Suspicious Macro Removal Windows policy setting. Customers remain protected through the Cloud Anti-malware for Microsoft Office Files settings. We refer customers to the FalconFlank Tech Alert in the CrowdStrike support portal.” 

As enterprises race to deploy AI coding agents, a new security challenge is emerging: organizations are creating high-privilege endpoint activity that many SOCs can’t actually see.

New research from the Abstract ASTRO team in a blog post that went live today examines telemetry from Anthropic’s Claude Code and Cowork and finds these tools create a rich but largely untapped detection source. It’s a source that can expose everything from shell execution and file access to plugin installs, MCP server interactions, and sensitive data leakage. The scary part? Most teams aren’t monitoring it.

ASTRO’s research also demonstrates how attackers could abuse AI coding workflows using techniques such as TrustFall, a recently disclosed flaw that can trigger arbitrary code execution simply through project trust prompts, potentially enabling credential theft, persistence, or data exfiltration.

A few findings and angles that may resonate with security readers:

• AI agents are becoming new endpoint hotspots with broad access across developer systems and applications
• Claude Code lacks native host telemetry, creating visibility and correlation challenges for SOC teams
• Organizations can reduce AI-agent log volume by 30–50% while preserving security visibility
• Detection opportunities include secret leakage, sensitive file access, malicious plugins, persistence attempts, and data exfiltration
• Researchers built a higher-fidelity detection approach correlating AI agent telemetry with EDR/process activity to reduce false positives

This speaks to a broader issue: security teams are entering an era where agent activity may need to be monitored the same way they monitor users, endpoints, and cloud infrastructure.

The Abstract ASTRO research team has just published a blog entitled: How Attackers Disable CloudTrail Without Calling StopLogging or DeleteTrail.

Security teams rely heavily on AWS CloudTrail as a source of truth for detecting breaches, but new research shows attackers can quietly disable or degrade logging without ever touching the APIs most defenders monitor.

In a new technical deep dive, ASTRO uncovers how adversaries are bypassing traditional detections (like StopLogging or DeleteTrail) and instead using lesser-known AWS APIs to blind logging systems while keeping them appearing fully operational.

Key findings that may interest your readers:

• Attackers can create “invisible activity zones” using PutEventSelectors, selectively excluding malicious actions from logs while CloudTrail continues to run normally.
• CloudTrail Lake can be silently neutralized via APIs like StopEventDataStoreIngestion and DeleteEventDataStore, halting or destroying long-term forensic visibility.
• Anomaly detection can be disabled outright by-passing empty parameters to PutInsightSelectors, removing automated detection of suspicious behavior.
• Critical guardrails can be dismantled through APIs like DeleteResourcePolicy and DeregisterOrganizationDelegatedAdmin, weakening cross-account protections.
• The real risk is in the sequence: individually, these API calls look like routine maintenance—but chained together, they allow attackers to erase evidence and evade detection entirely.

The research also outlines detection strategies, including how to identify subtle parameter changes and—more importantly—how to correlate multiple low-signal events into high-confidence alerts, something most SIEMs struggle to do.

This has major implications for DFIR teams and cloud security programs: organizations may believe they have full visibility, while attackers are actively operating in blind spots.

You can read the blog entry here: https://www.abstract.security/blog/how-attackers-disable-cloudtrail-without-calling-stoplogging-or-deletetrail

A new threat intelligence report from the Abstract’s Threat Research Organization (ASTRO) will reveal that the cybercrime economy has industrialized network breaches with specialized criminals now selling pre-compromised access to corporate networks for as little as $500.

Abstract’s report, “Priced to Move: The Underground Markets of Modern Cyberattacks,” examines the rapidly growing ecosystem of Initial Access Brokers (IABs): attackers who break into organizations and then sell that access to ransomware gangs and other threat actors.

Key findings from the research include:

1. Credential abuse is now the dominant entry point. 56% of incidents involved valid accounts without MFA.
2. Ransomware attacks surged 47% year over year, fueled by the growth of this underground access market.
3. Network access often sells for $500–$1,000, allowing attackers to target dozens of organizations simultaneously.
4. Median time from initial compromise to ransomware deployment has dropped to just five days.
5. Healthcare, government, and education are among the sectors seeing the fastest growth in IAB-driven attacks.

The economics are striking. The report details a healthcare breach where $2,200 worth of purchased access ultimately resulted in nearly $4 million in damage, a roughly 1,700x return on investment for attackers.

ASTRO says the rise of access brokers has fundamentally changed how cybercrime operates…turning network intrusions into a specialized supply chain where one group gains access, another sells it, and ransomware gangs monetize it.

You can read the research here:https://abstract.security/reports/priced-to-move

Researchers from Abstracts ASTRO research team have uncovered new developments in the evolving “Contagious Interview” campaign, showing how attackers are increasingly abusing developer tools—including Visual Studio Code and the AI coding editor Cursor AI Code Editor—to silently execute malware on developer machines. Here is the blog post: Contagious Interview: Evolution of VS Code and Cursor Tasks Infection Chains Part 2.

ASTRO analysts detail how attackers are embedding malicious commands into IDE task configuration files. When a developer opens a cloned repository and approves the standard workspace trust prompt, the tasks execute automatically which triggers multi-stage infection chains without requiring the victim to run code manually.

Key findings from the research include:

• New payload staging infrastructure: Attackers are shifting from previously exposed hosting platforms to GitHub Gists, URL shorteners, and Google Drive to stage malicious scripts and payloads.
• Developer-focused social engineering: Malicious repositories disguised as interview projects or legitimate development tools execute automatically when opened in an IDE.
• Multi-stage infection chains: Initial task execution downloads additional loaders and can ultimately deploy infostealers or backdoors targeting browser credentials, crypto wallets, and system data.
• Evasion tactics: Some payloads are hidden off-screen in configuration files or masquerade as legitimate GPU or driver tooling to avoid detection.

The report also outlines detection opportunities for security teams, including monitoring IDE-spawned shell commands, suspicious use of URL shorteners in configuration files, and unusual process chains involving Node.js and Python runtimes.

Given the growing use of AI-assisted development environments and the trust developers place in their toolchains, researchers warn this technique could become a major new software supply-chain attack vector.

The first blog post about Contagious Interview is here:: https://www.abstract.security/blog/contagious-interview-evolution-of-vscode-and-cursor-tasks-infection-chains.

Abstract’s ASTRO research team has just published a blog entitled: Critical Cisco Vulnerabilities: CVE-2026-20079 and CVE-2026-20131 Affecting Cisco Secure Firewall Management Center‍.

Earlier today, Cisco published several security advisories addressing vulnerabilities across its Secure Firewall product line. Two of these are rated critical with a CVSS score of 10.0 and affect Cisco Secure Firewall Management Center (FMC). Both can be exploited remotely by unauthenticated attackers to execute code on an affected device and obtain root access to the underlying operating system. Cisco has released software updates to address these vulnerabilities. Currently there are no workarounds for either vulnerability, making patching the only path to remediation. At the time of publishing, Cisco PSIRT is not aware of any public announcements or malicious use of these vulnerabilities.

This post covers the critical vulnerabilities in detail, along with a summary of additional high-severity issues disclosed in the same advisory bundle. Abstract also offers recommendations for immediate actions to take plus detection and monitoring bullets.

Abstract Security just published a blog this morning: Moving Laterally through Abuse of Managed Identities attached to VMs.  The blog was written by Abstract’s ASTRO research organization.

The research talks about how to put some detection for some type of managed identity abuse. Since managed Identities are very useful tools for the proper functioning of an Azure environment, it becomes difficult in case there are multiple resources attached to a single Managed Identity.

This can lead to the abuse of managed identities. Even though detection may vary depending on environment. For example, there might be some script which uses managed Identities to access other resources like another Virtual Machine. Therefore, this detection is very generalized form of detecting some type of managed identity abuse.

You can read the blog post here: https://www.abstract.security/blog/moving-laterally-through-abuse-of-managed-identities-attached-to-vms