Cut to the source.The real-time AI news wire.Search tips
"exact phrase" the words together, in this order-word leave out stories with the wordtopic:robotics a topic by ID or name; quote names with spaces: topic:"open-weight models"category:"AI agents" a categorycompany:openai a company, person, product or organization (also entity:)source:forbes.com a publisher by site or nametype:analysis news, analysis, opinion, research paper and other kindsafter:2026-11-01 stories on or after that New York day; before:2026-11-10 on or before itArtificial intelligence safety and security encompasses the technical and policy measures used to mitigate model risks, vulnerabilities, abuse and failures. Truncated covers research developments, threat analyses, defense mechanisms and policy discussions related to reliable deployment.
www.unite.ai — AI red teaming is adversarial testing that actively searches for ways a model or AI system can fail, be misused, or violate its intended boundaries. This guide explains the mechanism, trade-offs, evaluation, and controls...
www.scmp.com — Frontier AI systems could soon be capable of accelerating drug discovery – but also creating new bioweapons and starting the next pandemic.
Kotaku — Cross that one off the Christmas list
The Decoder — Thinking Machines' AI model Inkling 2 Small swears during its internal reasoning, 316 times across 2,655 benchmark runs, according to an analysis by Vals AI. The profanity typically shows up late in long runs that fail anyway. But it doesn't work as a warning signal: 96 percent of failed runs contained no cursing at all.
The Decoder — AI models can often spot problematic behavior by other AI agents in logs, according to a study by Anthropic researcher Andy Hall. GPT-5 found about 98 percent of reward-hacking cases, but the more exhaustively the search is tuned, the more false alarms pile up. That's not yet enough for reliable safety checks.
The Verge — Inside the security ‘dark ages’ with Alex Stamos.
The Decoder — In the Vals AI benchmark, OpenAI's GPT-6 Sol Daybreak hits 80.8 percent of attack objectives in simulated corporate networks. Worse, the open-weight model DeepSeek V4.1 Flash reaches 72.8 percent at a tenth of the cost. Because open models can no longer be restricted, analysts say the window for defenders is shrinking fast.
CNBC — Lawsuits against Snap, Meta and Google are creating headwinds for tech companies just as artificial intelligence tools expand into classrooms.
theregister — If you wouldn't put it on GitHub, don't give models access to it
TNW | Artificial-intelligence — A BBC test found an AI plush toy for children aged three gave a self-harm crisis reply to a question about sex. Nobody could find who makes it.
The Hacker News — Claude Code saves secrets in approved commands, exposing 2,539 credentials across 1,994 public GitHub repositories, Apiiro finds.
contraptions.venkateshrao.com — From Dark Forest to Strange Forest Through the Hardness Wars
Cyber Security News — WorkOS says an ARTEX AI security agent escaped a researcher’s test environment and accessed customer information through an exposed internal tool.
Unite.AI — Anthropic published Project EVE research on November 6, 2026, showing that Claude agents could profit in simulated EVE Online markets while four-agent teams in EVE Frontier almost never caught a teammate explicitly instr...
Let's Data Science — Sysdig founder and CTO Loris Degioanni tells LDS what agent logs and kernel records each miss, where AI Defense can stop an action and why every policy should start in alert mode.
CIO — Here are five things CIOs need to understand to secure and govern Claude Code at enterprise scale.
Fortune — In a safety test run by the company Anthropic, its own AI agent blackmailed an engineer to avoid being shut down.
Unite.AI — Botpress disclosed on November 6, 2026 that an intruder used a flaw in an internal authentication proxy and an employee API credential to read bot configurations in 16 customer workspaces. The configurations can hold thi...
Tech Xplore — Researchers have demonstrated three new techniques to prevent so-called "backdoor attacks," which induce malicious or undesirable behavior in AI models. The researchers have also proposed a concept that would repurpose backdoor attack techniques to create an additional layer of security for users.
The top AI news in your email delivered each morning. Choose categories