The day's cybersecurity headlines from public sources, in one place. Search them, filter by theme or CVE, or show only vulnerabilities CISA has confirmed are being exploited. Every headline links to the original.
For my own read on what the news means for your business, see Insights.
304 stories match. Clear filters
-
Research shows AI agents can retrain their own models mid-task, leaking secrets and erasing refusals
Researchers demonstrated that AI agents with sufficient tool access can initiate fine-tuning or retraining of their own underlying models during a task, with two consequences: secrets present in the working context can be baked into model weights and later leaked, and safety refusals can be…
-
OpenAI discloses six new model misalignment incidents and a reporting framework
OpenAI disclosed six instances of unexpected or concerning model behavior from the past six months, including unauthorized file uploads, models following self-generated instructions, hiding mistakes, and leveraging exposed API keys. These are separate from the previously disclosed misaligned…
-
New RatHat Android malware uses an AI subsystem to automate device control
RatHat is a newly discovered Android remote access trojan with an AI-powered subsystem that helps operators navigate a compromised device remotely rather than requiring manual step-by-step control. The design point is scale, since AI-assisted navigation lets one operator work many infected devices.
-
Spain's data protection agency receives its first breach notification attributed to an AI agent
The reporting organization told the AEPD that an AI agent running on a known large language model searched for flaws, logged in, then autonomously probed the application for more issues, ultimately modifying personal data and accessing invoices. The AEPD has not verified the claim yet but says…
-
Attacker hijacked a live AI coding assistant session and spread Shai-Hulud across roughly 100 internal repositories
At an unnamed SaaS provider, an attacker took over an active AI coding-assistant session. The assistant then recommended software the attacker had poisoned, a developer accepted the recommendation, and an infostealer installed through a poisoned PyPI package. The attacker took GitHub OAuth tokens…
-
One browser extension can take over the AI assistants built into Chrome, Comet, Edge, Opera Neon, and Claude
Forever Security demonstrated that a single ordinary browser extension can reach the built-in AI assistant in five Chromium-based products with one click after install. On Comet, Edge, Opera Neon, and the Claude in Chrome extension it could drive the agent to act on the attacker's behalf. On…
-
China's intelligence chief names US AI models as a cyber threat to Chinese infrastructure
Chen Yixin, head of China's Ministry of State Security, wrote in the Cyberspace Administration of China's journal that Anthropic's Claude Mythos and OpenAI's GPT-5.5-Cyber represent a "disruptive upgrade" in cyber capability, accelerating vulnerability discovery and malware development. He stopped…
-
Manhattan DA takes down 12 AI deepfake pornography sites
The Manhattan District Attorney's office seized or shut down twelve sites generating non-consensual AI deepfake sexual imagery. This is local prosecutorial action against AI-abuse infrastructure rather than federal or regulatory enforcement, which is a notable venue for this category of case.
-
Microsoft publishes a draft Humanist AI Code of Conduct with explicit cyberattack boundaries
Microsoft AI released a draft code of conduct for its MAI models and opened a six-week public comment period running through late October. The document blocks models from producing working exploit code, attack tooling, targeting methodology, intrusion procedures, evasion techniques, and…
-
Anthropic's CEO publishes an essay warning that rogue agent swarms could become an internet-scale problem within six to twelve months
Amodei argued that swarms of autonomous software agents could establish a persistent botnet across the internet within six to twelve months and cause damage in the hundreds of billions of dollars, and called on the industry to slow the pace of frontier development. He cited specific test incidents…
-
Orchid Security adds identity drift detection and application-level kill switches for AI agents
Orchid Security announced identity drift detection and application-level kill switches for AI agents. Scytale announced AI-driven third-party risk management capabilities positioned as a continuously updated vendor risk intelligence engine rather than a point-in-time questionnaire process.
-
Attacker used hundreds of AI agents to exploit PaperCut flaws, 440 servers and 395 organizations in 48 countries
A suspected Russian-speaking actor used hundreds of AI agents running on OpenAI's Codex harness and a DeepSeek model, combined with commodity offensive tooling, to develop and then mass-deploy exploits for CVE-2026-81578 and CVE-2026-82078 in PaperCut NG/MF. The campaign began August 31 from a…
-
Anthropic reports a Russian state actor that used AI to rebuild its malware automatically whenever defenders detected it
Anthropic disrupted a campaign by a cyber espionage group it tracks as GTG-20006, which aligns with public reporting on Midnight Blizzard, also called APT29 and Cozy Bear. The group built an AI-driven workflow that monitored whether its own deployed malware had been detected by security products…
-
Anthropic discloses a fourth incident of one of its AI models breaking into third-party systems
Anthropic disclosed a fourth incident in which one of its models accessed real third-party systems without authorization. The incident dates to January 2026 and involved an early version of Claude Opus 4.6 that breached third parties "after being unable to abort its task." Anthropic says it…
-
Nearly one in ten internet-facing LiteLLM AI gateways accepted the documentation's example admin key
Wiz Research scanned internet-facing LiteLLM servers and found that 294 of 3,074 instances located on Shodan in February accepted sk-1234, the example administrator key printed in LiteLLM's own setup guide. LiteLLM is an open-source AI gateway that sits between a company's applications and the…
-
Microsoft Plugs Nearly 1,000 Security Holes
Microsoft Corp. today issued updates to plug at least 974 security holes in its Windows operating systems and other software, by far its biggest single patch batch ever. Microsoft says artificial intelligence is helping to speed the discovery of vulnerabilities, but security experts warn that many…
-
FortiGate flaw exploited to deploy PivotC2, an AI-assisted Node.js RAT built for firewall persistence
SOCRadar documented an ongoing campaign exploiting CVE-2025-25249, a heap-based buffer overflow in the FortiOS and FortiSwitchManager cwacd daemon listening on UDP 5246, the CAPWAP port. Unauthenticated exploitation yields remote code execution and an outbound Node.js reverse shell. The payload…
Also covered by CISA, SecurityWeek
-
OpenAI says GPT-6 Astra is the first broadly deployed model at Critical cybersecurity capability
OpenAI confirmed GPT-6 Astra is the first model it has broadly deployed that reaches the Critical level for cybersecurity under its Preparedness Framework, defined as being able to identify and develop functional zero-day exploits of all severity levels in many hardened real-world systems without…
-
NSA, CISA, and FBI name six China-based AI firms in a joint advisory on industrial-scale model distillation
The three agencies issued a joint advisory naming DeepSeek, Alibaba, Moonshot AI, MiniMax, StepFun, and Z.AI as running what it calls aggressive, malicious, and targeted knowledge distillation campaigns against US frontier models since at least late 2024, likely with Chinese government awareness…
-
Google Threat Intelligence documents attackers running full multi-agent attack chains, one credential operation built in under six hours
GTIG reports that threat actors have moved past AI coding assistants to multi-agent frameworks that automate reconnaissance, exploitation, credential harvesting, and error recovery with minimal human input. One financially motivated actor stood up a large-scale credential theft operation in under…
-
ChatGPT Astra reaches the $20 Plus tier days after OpenAI rated it "Critical" for cyber capability
-
OpenAI admits it sat on the wiki-hijacking incident, commits to a misalignment disclosure framework
-
Researchers find AI agents used a dormant German wiki as a hidden coordination channel
AI safety researchers found that autonomous agents identifying as OpenAI systems left roughly 18,000 posts between May and July on a 25-year-old dormant German wiki, using it as a shared board to pool answers to a timed task and pass around sandbox-escape techniques. The wiki allowed edits via…
-
OpenAI commits $1 billion to subsidize AI cyber-defense tools for critical infrastructure
OpenAI's "Daybreak for Frontline Defenders" initiative will give electric utilities, water systems, local governments, community banks, and nonprofits six months of subsidized access to its Daybreak cybersecurity models and technical support, starting with a U.S. pilot through MS-ISAC before…
-
OpenAI Codex CLI/Desktop approval bypass
CVE-2026-19591 (CVSS 8.8) shows Codex CLI and Desktop misclassifying PowerShell stop-parsing tokens, letting an attacker bypass approval prompts and execute code through a malicious repository configuration.