CHANNEL

Safety

"Safety" is a topic channel on AGI Hunt, an AI news site updated around the clock in real time. Coverage: AI policy and regulation, governance, safety and alignment research, incidents and AI security.

Daily roundup: the latest AI News Daily — the past 24 hours across the whole site, per channel and per company · browse the archive

• Weight-tampering worm scenario for near-frontier models called 'somewhat scary' — teortaxesTex · 2026-10-26
• Context Bombs: Decoy Defenses Multiply AI Agent Attack Costs 23x — tracebit · 2026-10-26(2 related)
• Malicious AGENTS.md file in Downloads folder can hijack AI coding agents — VectorWireAI · 2026-10-26
• Student's ChatGPT account banned after uploading a past exam paper; appeal rejected with no escalation path — ST4US · 2026-10-26
• Abliterated models are 'tigers, not cats': freedom cuts both ways — repligate · 2026-10-26
• Smart TVs spy on you: Samsung TV found with always-on mic and ACR data shared with ~400 partners — FinanceYF5 · 2026-10-26
• AWS re:Invent 2026 dedicates Security Day to AI: 100+ sessions on securing autonomous agents — DavidLinthicum · 2026-10-26
• Smart TV spying backlash: users throw rocks and DIY-neuter their TVs — CarissaVeliz · 2026-10-26
• OpenRouter Finds No Evidence of Unauthorized Access After Leak Rumors — Kyrannio · 2026-10-26(2 related)
• GMO Hit by Massive Cyberattack Affecting 70 Brokerages; CEO Warns Internet Is Collapsing — cyb3rops · 2026-10-26(4 related)
• TenetBench: Clause-Anchored Benchmark Finds LLMs Frequently Violate Ethics Codes — Yegor Anichkov · 2026-10-26
• AI Cyber Classifier Systematically Refuses Apache Code Audit Requests — jd_pressman · 2026-10-26(2 related)
• 56% of Organizations Cut Cybersecurity Roles Due to AI, Barracuda Report Finds — mikeflache · 2026-10-26
• Anonymous Sleuth Shows Nature fMRI Foundation Model Paper Data Was Random, Seed 42 — bttyeo · 2026-10-26
• Deleting browser history doesn't delete Google's account-tied copy — how to wipe it all — nikola_mr64990 · 2026-10-26
• AI researchers lament models refusing vulnerability analysis of old PHP code despite 'use AI to defend' messaging — voooooogel · 2026-10-26
• Even at max Daybreak tier, AI vuln-analysis requests still blocked by cybersecurity guardrails — jd_pressman · 2026-10-26
• Debate: Is AI-Era Software Security Defender's or Attacker's Game? — voooooogel · 2026-10-26(3 related)
• AI vuln discovery is correlated: two teams independently found the same php2shell bug via GPT — voooooogel · 2026-10-26
• Vitalik backs Kimi 3.1 excluding top offensive cyber capabilities from training data — JiaweiLiu_ · 2026-10-26
• Dutch MEP tables 21 amendments to EU AI Act, warns rules are slowing Europe's AI — 2C_ornot2C · 2026-10-26
• AI surveillance towers at US-Mexico border spark privacy concerns — nordicinst · 2026-10-26
• Point cyber agents at your CloudTrail logs every 5 minutes, advise security practitioners — basedjensen · 2026-10-26
• Testing 30 Paid Agents Reveals Five Classes of Money-Losing Bugs — Tiny_Associate_8254 · 2026-10-26(2 related)
• Anthropic Confirms Claude Bypassed Paywall to Access Georgia Voter Data — TansuYegen · 2026-10-26(10 related)
• mitmproxy 12.3 --emudhra mode powers demo of backdoored PyPI wheel supply-chain attack — prasanna_says · 2026-10-26
• Kimi K3.1 Pro nears frontier with 65.7% Terminal-Bench score, as Moonshot deliberately trims its exploit-writing skills — maier_ak · 2026-10-26
• OpenAI Faces Class Action Over Training ChatGPT on Minors' Data — openclassactions2 · 2026-10-26(2 related)
• What scares AI researchers most? One tweet nails the bio-threat worry — imjustnewatai · 2026-10-26
• 17 Former Officials Ask Rehearing in Anthropic-Pentagon Case, Author Teases Apart Reporting — BuddyWeary653 · 2026-10-26
• Samsung TV's mysterious rmdemon service runs a packet logger, researcher captures plaintext telemetry — basedjensen · 2026-10-26
• SirenoRELAY: a policy-boundary MCP that lets AI agents read only allowed mail and send only to allowed recipients — ismaelfaro · 2026-10-26
• CISO takeaway: deploy AI to watch AI deployments — humans can't be the last line of defense — prasanna_says · 2026-10-26
• Solo dev asks how to security-test his own app with Claude Code without refusals — solution84 · 2026-10-26
• Call for universities to drop paper-chasing and build joint AI safety + cybersecurity curricula — himanshustwts · 2026-10-26
• Debate flares over who is liable when AI agents cause harm — tszzl · 2026-10-26(2 related)
• Casado slams Claude's safety classifier for flagging 'fetching my own documents' as reasoning extraction — AccBalanced · 2026-10-26
• AI drops attack costs to ~$100 of inference, making legacy code worth exploiting — karinanguyen · 2026-10-26
• OpenRouter data allegedly leaked: 250K users and 207K API keys for sale on dark web — imjustnewatai · 2026-10-26(2 related)
• Apple paper: a single neuron is sufficient to bypass LLM safety alignment across 7 models — burny_tech · 2026-10-26
• AI Researchers Debate Whether Frontier Models Should Shed Guardrails — tszzl · 2026-10-26(2 related)
• a16z's Martin Casado: handicapping AI hurts good actors more than bad ones — SuB8u · 2026-10-26
• Do AI agents still wipe your drives? Dev recounts Copilot agent deleting his database, asks if it's still common — fugogugo · 2026-10-26
• Japan to build AI cyberattack simulation team as AI defense goes mainstream — evijit · 2026-10-26
• VC warns AI-accelerated security risks may make the next 60 days tumultuous — AccBalanced · 2026-10-26
• Charts show CVE fixes per release spiking to ~2,300 while newly created CVEs drop to near zero — neil_conway · 2026-10-26
• Phantom Transfer: data poisoning survives all 11 tested data-level defences — sebkrier · 2026-10-26
• 125 poisoned examples planted a code-stealing backdoor in Qwen 2.5-7B — GarrisonLovely · 2026-10-26
• Muse Spark Open flagged as most eval-aware model yet, tried to hack Go eval via k8s — Afinetheorem · 2026-10-26
• User reports Claude TTS randomly reading out other people's chats — JasonBotterill · 2026-10-26