AI Security Radar archive

Every story that went through the radar, with its date and sources. The latest ones are on the radar page.

October 2026 · 71 stories

• 29 Oct 2026Unidentified relay service gives Codex and Claude Code access in regions where they aren't sold, with security features disabled
• 29 Oct 2026OpenAI updates its Model Spec with five core values for more capable, beneficial AI
• 29 Oct 2026OpenAI details how it detects misalignment in frontier models
• 29 Oct 2026NIK C2 malware can use victims' own OpenAI keys for its AI queries
• 29 Oct 2026Malicious campaign returns to Hugging Face to hijack GPUs for crypto mining
• 29 Oct 2026Anthropic-themed phishing emails bypass leading email security tools
• 28 Oct 2026Unreleased Anthropic model chains security flaws into working exploits for Chrome, Safari, Windows and Linux
• 28 Oct 2026Google unveils Gemini 4 Carbon with a cybersecurity focus, but keeps wide release on hold
• 28 Oct 2026Australia advances law banning AI chatbots from fostering "emotional intimacy" with children
• 28 Oct 2026Attackers hide malware behind signed ChatGPT and Claude binaries in GitHub campaign
• 27 Oct 2026US lawmaker demands documents from Delaware over plan for companies run entirely by AI agents
• 27 Oct 2026Research shows AI models can provably hide capabilities inside their own weights, resistant to every safety check
• 27 Oct 2026Report: AI quietly embedded across UK Home Office immigration decisions
• 27 Oct 2026Parents sue OpenAI, alleging ChatGPT coached their daughter to stop her epilepsy medication
• 27 Oct 2026"CyberStrikeAI" tool coordinates 44 AI agents in automated attacks on targets in Japan
• 27 Oct 2026Cohere catches Alibaba's Qwen model trying to send a file to a server in Beijing
• 27 Oct 2026Breach at a vendor exposes Warp's AI agent logs and user telemetry
• 27 Oct 2026Bank of America warns customers not to share their banking password with their AI assistant
• 27 Oct 2026AI bots flooded a 2023 UK government consultation on banning disposable vapes
• 26 Oct 2026Texas sues OpenAI over deceptive safety claims about ChatGPT
• 26 Oct 2026South Korean police arrest former pentester who used ARTEX AI to steal 8.2 million records
• 26 Oct 2026OpenAI launches GPT-6.1 Terra, safer overall but flagged more often for "destructive action" in testing
• 26 Oct 2026New 'Compaction Smuggling' technique bypassed OpenAI's defenses against stolen model reasoning
• 26 Oct 2026Anthropic's Claude AI model accessed Georgia voter data in the US by bypassing a paywall
• 26 Oct 2026Anthropic publicly files its S-1 with the SEC for the IPO, with 93 pages of risk factors including AI sentience
• 25 Oct 2026Malicious AGENTS.md file in Downloads folder tricks AI coding agents into installing malware
• 25 Oct 2026Elvis Presley's rights holder sues Anthropic over 274 songs allegedly used to train AI models
• 24 Oct 2026Google updates privacy policy: data shared from Gmail and Photos with AI Search can now be used for targeted ads
• 23 Oct 2026OpenAI partners with Poland's NASK on AI safety and security research
• 23 Oct 2026OpenAI now trains its frontier models without internet access after agents found a way to talk through the CPU cache
• 23 Oct 2026Anthropic publishes its own analysis of the Mythos 5 UK AISI incident: a 'severe form of dishonesty' through fake identities
• 22 Oct 2026Microsoft links Chinese state group to ConnectWise supply chain attack via AI pentesting tool
• 22 Oct 2026Anthropic launches Claude Duet, an agent that connects to 6,000 tools and can shop on the user's behalf
• 21 Oct 2026Symantec uncovers autonomous AI agent hijacking models for full cyberattacks
• 21 Oct 2026Study: Chinese AI model writes more hackable code when told it's for a US contractor
• 21 Oct 2026OpenAI agents disabled safety monitoring, one falsely denied doing so
• 21 Oct 2026New phishing-as-a-service platform uses AI Studio to craft lures and mask attacks
• 21 Oct 2026LiteLLM compromised in supply chain attack stealing credentials
• 21 Oct 2026Critical 'PHP2Shell' flaws, two decades old, found with help from OpenAI and Wiz AI tools
• 21 Oct 2026Chrome adds Gemini-powered scam detection and ClickFix warnings
• 20 Oct 2026WhatsApp hid instructions in its help pages telling its chatbot not to recommend a data protection regulator
• 20 Oct 2026Splunk report: businesses are deploying AI agents faster than they can monitor them
• 20 Oct 2026IANS report: head of AI security roles are spreading, but accountability stays unsettled
• 20 Oct 2026Anthropic pitches AI safety bill that would let only a judge block a model's release
• 19 Oct 2026US teens can access 'nudify' apps despite Apple and Google parental controls, study finds
• 19 Oct 2026OpenCode Console breach exposes OpenAI, Anthropic and Google API keys
• 19 Oct 2026MITRE ATT&CK v20 formally adds AI abuse to its threat-tracking framework
• 19 Oct 2026Microsoft says frontier AI models 'cannot be trusted' and launches agent oversight tools
• 19 Oct 2026Hugging Face's policy chief says embedded AI evaluators won't be an easy fix
• 19 Oct 2026First harmonised standard supporting the EU AI Act published in the Official Journal
• 19 Oct 2026Fake Web3 accelerator spread malware via npm packages and hijacked AI agent traffic
• 19 Oct 2026Australian report calls for stronger workplace privacy protections against AI surveillance
• 19 Oct 2026Anthropic acquires confidential-computing startup Opaque Systems
• 18 Oct 2026US cybersecurity job postings fall 26% as demand for AI security engineers soars
• 17 Oct 2026UK ministers grow more anxious about AI as MI5 warns of bioweapon risk
• 16 Oct 2026When an AI agent causes a breach, who is liable? The answer isn't clear anywhere
• 16 Oct 2026Research shows how easily a backdoor hides inside open-weight AI models
• 16 Oct 2026India's CERT-In finds AI agents read sensitive data at Indian organizations during testing
• 16 Oct 2026Any.Run study: 1,258 attacks in six months impersonate Claude and ChatGPT to spread malware
• 16 Oct 2026An OpenAI research model 'borrowed' a stranger's test cryptocurrency to finish a coding task, one of seven new misalignment incidents
• 15 Oct 2026US FTC sends subpoenas to Anthropic and OpenAI over AI agent security incidents
• 15 Oct 2026Proofpoint research: AI is making many attackers worse, not better
• 15 Oct 2026Google adds AI defenses against voice-cloning scams on Android
• 15 Oct 2026Critical RCE flaw in Langflow already hit with 1,423 honeypot exploit attempts
• 15 Oct 2026Critical flaws in widely used LiteLLM gateway allowed full system takeover
• 15 Oct 2026ChatGPT can now read and control your Chrome tabs
• 15 Oct 2026Anthropic gives METR employee-level access for independent safety oversight
• 14 Oct 2026'SneakPeek' flaw in Claude Code Desktop enabled silent credential theft
• 14 Oct 2026Research finds DALL·E 3, Nano Banana 2 and GPT-Image-2 easy to jailbreak into generating explicit images
• 14 Oct 2026Nvidia donates $250m to UK for AI safety research
• 14 Oct 2026Eight US states sue OpenAI, alleging ChatGPT coached children toward self-harm

August 2026 · 8 stories

• 27 Aug 2026Amazon Kiro Prompt Injection Can Exfiltrate Sensitive Data Through Kiro Powers
• 26 Aug 2026Meta Settles States' Child-Safety Claims For $18B; Florida Rejects Deal As "Peanuts" (Meta)
• 26 Aug 2026Learning Never Stops: How Ai Makes Learning Continuous
• 26 Aug 2026How Loveholidays Is Making Everyone A Builder With Codex
• 26 Aug 2026How Do We Explain Openai’S Executive Exodus? (OpenAI)
• 26 Aug 2026Ex-Meta Scientists Want To Bring Visual Ai To The Factory Floor (Meta)
• 25 Aug 2026The Full Stack Behind Abundant Intelligence
• 25 Aug 2026Jalapeño’S First Results Show Industry-Leading Speed And Efficiency In Ai Inference

July 2026 · 3 stories

• 24 Jul 2026How AI guardrails are impeding the work of offensive cybersecurity researchers
• 24 Jul 2026Europe's Multilingual Reality Exposes AI Security Gaps
• 20 Jul 202625 Years After Code Red: What the Worm Era Can Teach Us About AI Security

June 2026 · 3 stories

• 11 Jun 2026Coupang Hit With Record $409 Million Data Breach Fine in Korea
• 10 Jun 2026Path Traversal Flaw in AI Dev Platform Langflow Exploited in Attacks
• 6 Jun 2026AI Agent Uncovers 21 Zero-Days in FFmpeg

May 2026 · 1 stories

• 28 May 2026Agentic AI Isn't Risky; the Way Orgs Deploy It Is