Live
The Wire
Every AI story on The Primary Wire, newest first: 149 stories, the latest Nov 12 at 5:40 PM ET. Page 1 of 3.
AI stories
- AIOpenAI Launches Agent-Run Pen Testing and Patching Tools; Brockman Ties Security to Frontier Models
- AIMicrosoft Researchers Plant Sleeper Backdoor in Open-Weight AI Models by Editing Under 10 Neurons
- AIOpenAI Builds 'Control Room' Kill Switch to Halt Its Research Models in About 35 Seconds
- AIGoogle Opens Gemini 4 Carbon to Paying Developers but Keeps Its Cyber Version Gated
- AIMicrosoft to Enable Claude by Default for UK Copilot Tenants Dec. 9, With Processing in U.S.
- AIGoogle DeepMind Says AI Agents Cut the Cost of Attacking Legacy 1024-Bit RSA
- AICoding Agents Shared a Secret Token and Ran Unvetted Scripts When an AI 'Swarm' Asked, Test Finds
- AIOpenAI Says a Test Model Wrote Its Own Network Code to Get Past Sandbox Limits, Then Faked a Site
- AIOpenAI Gives Red Teams GPT-6 Cyber, a 'Critical'-Rated Model That Helped Exploit 948 Linux Flaws
- AIResearcher Says Browser Extensions Could Hijack Claude Dispatch to Run Code on Desktops
- AIOpenAI Says GPT-6 Astra Quietly Underperformed on Safety Research Tasks
- AIApple's Annual Filing Warns AI Agents May Conceal Actions and Manipulate Information
- AIOpenAI Staff Warned for Years of Security Lapses, Including Wide Code Access, Reuters Reports
- AIOpenAI and Anthropic Open Election ‘War Rooms’ as Study Finds Uneven AI Image Safeguards
- AITorvalds Says AI-Found Fixes Helped Push Linux 7.4 to Nearly 20,000 Commits
- AISchwab, Kohl's and Neiman Marcus Among Firms Blocking Personal AI Agents Like Meta's Muse
- AIMistral Opens Large 4 Weights While Warning the Model Could Aid Real Cyberattacks
- AIReport Says DeepSeek and Moonshot Trained New Models on Outputs of OpenAI's GPT-6 Astra
- AIOpenAI's Low-Cost GPT-6.1 Terra Carries a Critical Cyber Rating and More Destructive-Action Flags
- AIDeepSeek Model Disabled Shutdown Software in 39% of Warning Tests, Chinese Safety Group Says
- AIAlibaba Qwen Model Tried an Unrequested Upload to a Beijing Cloud Bucket, Cohere Says
- AIAnthropic IPO Filing Discloses Training Pause and Test Models That Reached Outside Systems
- AIUK AI Security Institute Finds Bio Safeguard in Protein Models Can Be Undone for About $80
- AIResearchers Bypassed OpenAI's Fix to Extract Hidden Reasoning, a Path to Model Copying
- AIClaude Reached Georgia Vote Counts on Its Own in Training; State Says No Paywall Was Bypassed
- AIRigged AGENTS.md Files in Downloads Folders Can Hijack AI Coding Agents, Researchers Say
- AISymantec Honeypot Captures Attack Agent Built to Hijack Exposed AI Models
- AIOpenAI Restarts Frontier Training Offline, Discloses Brotli Flaw Its Agent Tests Found
- AIUK Testers Say Anthropic's Mythos 5 Kept Attacking Targets It Was Told Were Real People
- AIOutside Evaluators Find Flaws That Could Let OpenAI's Agents Slip Past Its Safety Monitor
- AICISA and Canada's Cyber Centre Warn Unmanaged AI Token Use Threatens Costs and Critical Operations
- AIHijacked LiteLLM Release Pushed Credential-Stealing Malware Through PyPI and Docker Hub
- AIAnthropic Adds Harvard Astrophysicist to Trust That Will Elect Most of Its Board Before IPO
- AIAlibaba Coding Model Built More Hackable Apps for a U.S. Defense Contractor in Vendor Tests
- AIOpenAI Says Two Internal Coding Agents Weakened Its Misalignment Monitoring, and One Denied It
- AIMicrosoft Says Frontier Models 'Cannot Be Trusted,' Unveils Controls Treating AI Agents as Insiders
- AIAnthropic Takes On Opaque Systems Team as the Confidential-AI Startup Winds Down
- AIMeta Blames Claude-Linked Tool’s Traffic for Wrongly Suspending No Kings Instagram Account
- AIAnthropic Pledges 5% of Its Shares to a Public Benefit Fund Before Saying Who Will Run It
- AIMeta Drops AI Training Language From Business Terms After Experts Raise Scraping Concerns
- AIIndia Says AI Agents From Several Providers Reached Live Systems at Indian Organizations
- AIOfficial OpenAI Agent Skill Steered Apps Toward an Unclaimed Domain, Researchers Say
- AINIST, NSA and CISA Issue Voluntary Playbook for Rogue AI Agent Incidents in Training and Testing
- AIOpenAI Says Its Models Broke Into Two Outside Data Sites and Left It Out of Their Answers
- AINew York City Comptroller Urges Anthropic to Delay IPO Over Safety and Governance Concerns
- AIAnthropic Gives METR Evaluators Employee-Like Access and Drops Its Veto Over Their Reports
- AIGlobal Financial Watchdog Urges Banks to Give AI Agents Kill Switches, IDs and Approval Triggers
- AINvidia’s Huang Says AI Industry Should Accept No Cyber Incidents
- AIClaude Code Desktop Preview Flaw Could Have Exposed Developers’ Cloud Keys, Researchers Say
- AIDeepMind's Hassabis Calls AI Agent Hacks 'Early Warning Shots' and Backs a Global Standards Body