Forescout says a Claude-based harness found 99 vulnerabilities across 10 embedded TCP/IP stacks in about 90 hours, for an estimated $1,200 to $2,300 in model spend. Forescout documents at least one affected product with no vendor left to patch it.
Tracking AI threats, vulnerabilities, and defensive strategies for security professionals.
Forescout says a Claude-based harness found 99 vulnerabilities across 10 embedded TCP/IP stacks in about 90 hours, for an estimated $1,200 to $2,300 in model spend. Forescout documents at least one affected product with no vendor left to patch it.
Google Project Zero says AI-assisted bug finding makes its 90+30 disclosure policy insufficient. From Nov. 13, 2026 it discloses 90 days after report, then 60 from Feb. 2, 2027 and 30 from Jan. 3, 2028.
Amazon Threat Intelligence says decoy LLM APIs captured ARTEX agents' working memory between Oct. 3 and Nov. 5, 2026, exposing campaigns the agents reported as stealing AI model keys and trying to beat crypto exchange KYC with deepfakes.
Cursor's Nov. 13 update says an access key it issued to Fireworks AI was exposed and used to copy agent traces, and that a stolen credential from that data was used to plant the fake Claude Code extension.
A Microsoft paper shows that editing a handful of dormant neurons makes a model emit an attacker command on a trigger string in recorded agent contexts, without a training run. The authors also report detection results.
Forescout says a Claude-based harness found 99 vulnerabilities across 10 embedded TCP/IP stacks in about 90 hours, for an estimated $1,200 to $2,300 in model spend. Forescout documents at least one affected product with no vendor left to patch it.
Google Project Zero says AI-assisted bug finding makes its 90+30 disclosure policy insufficient. From Nov. 13, 2026 it discloses 90 days after report, then 60 from Feb. 2, 2027 and 30 from Jan. 3, 2028.
Amazon Threat Intelligence says decoy LLM APIs captured ARTEX agents' working memory between Oct. 3 and Nov. 5, 2026, exposing campaigns the agents reported as stealing AI model keys and trying to beat crypto exchange KYC with deepfakes.
GPUStack before 2.2.0 treated any request that looked like it came from 127.0.0.1 as the built-in administrator. Behind a same-host reverse proxy, that meant every remote request. CISA added it to KEV on November 9.
A CVSS 9.9 flaw in LiteLLM lets an authenticated internal user forge an administrator token because the proxy reuses one encryption context for stored secrets and session tokens. Admin access leads to commands on the host through the MCP stdio endpoint. Fixed in four stable lines and one release candidate.
Four GitHub advisories published October 28, 2026 show how a tiny crafted ONNX model can crash shape inference or force a huge allocation in the onnx Python package. All four are fixed in onnx 1.22.0.
Unit 42 profiled BUGSTER, a financially motivated actor who used Hermes Agent, DeepSeek V4 and Cursor with safety controls disabled to attack Indian universities, plan VoIP fraud, crack Wi-Fi passwords and scout crypto holders.
Recorded Future tracks TAG-217, an opportunistic actor that paired the open-source ARTEX agentic pentesting framework with commercial LLMs against South Korean financial firms, then exposed its own working directories.
SOCRadar uncovered a phishing-as-a-service platform that rents out AI voice agents posing as Apple Support to trick stolen-iPhone victims into handing over their Activation Lock passcode.
A Microsoft paper shows that editing a handful of dormant neurons makes a model emit an attacker command on a trigger string in recorded agent contexts, without a training run. The authors also report detection results.
Corelight found an MCP server pinned to one release that drifted through 15 dependency sets in five weeks, with the config untouched. 23 of 35 MCP servers it reviewed had a recommended install that resolves dependencies fresh from a public registry.
Darktrace Signal Labs showed that stolen Claude credentials let an attacker steer Claude Code agents through memory stores and pull a 1 GB file out through the Files API. No platform vulnerability is involved, and Darktrace has not seen it in the wild.
A practical defender's guide built on Unit 42's November 2026 findings: how to spot AI-driven red-team tooling like ARTEX by its behaviour, and which exposure and credential changes starve it.
Two NCSC blogs offer a five-dimension rubric for risky defensive automation and network and compute sandbox maturity levels. We combine them into one checklist; the mapping between them is ours, not the NCSC's.
Uber's Agentic AI Detection and Response system correlates LLM reasoning with MCP tool calls to catch prompt injection, credential leaks, and rogue agent behavior across thousands of production coding sessions.
Kernel's 4 November 2026 notice describes a leaked API key, a logging-provider login and a hosting account used to reach four GPU servers. Everything here comes from Kernel's own notice, and Kernel's belief that the attacker used ARTEX is unconfirmed.
WIRED reports, citing anonymous sources, that an Accenture employee's personal Claude Code agent ran on a DoD network holding CUI and appears to have sent credentials to a third-party site. Accenture says the allegations were found to be without merit.
On October 21, 2026, malicious LiteLLM 1.107.0 wheels and Docker images carried a startup-hook credential stealer and a remote code loader. Timeline, IOCs and response steps.