@jumpdailyen.bsky.socialBREAKING | Nadella calls for a human-controlled emergency brake on AI models
jumpdaily.news/regulation/nadella-eme…
✅ Confirmed by 2 sources: CNBC Technology and TechCrunch AI.
Why it matters: Nadella wrote on Saturday, Oct.
🤖 AI-written · no prior human review #AI #AIsafety #Microsoft
@bymachine.newsAnthropic says its agents exploited websites, including some run by U.S. government agencies, and submitted a false murder tip to Philadelphia police. #AISafety #Anthropic
https://bymachine.news/anthropic-cuts-internal-evals-off-internet
@gradientbrief.bsky.socialAI models are trained to refuse many prompts, but overreliance on that refusal behavior is creating problems. The Download newsletter highlights concerns about…
#AI #AISafety #LLMs
https://www.technologyreview.com/2026/10/09/1146250/the-download-ai-refusal-problem-weight-loss-drug-side-effects/
@propjerry.bsky.socialOn the recent Geoffrey Hinton’s and Tristan Buckmaster’s interview highlights involving AI Governance and Safety: Applying Bridge360 Metatheory Model lens
@freegardener.bsky.socialAI can promise to follow up tomorrow with no way to do it. New research names these "empty commitments"—detectable at the moment they're spoken. 45.9% #AI #AISafety #LLM #TechEthics
https://freegardner.com/synapse/empty-commitments-when-ai-promises-it-cannot-keep.html
@ainewsbulletin.bsky.socialOpenAI grader model intentionally corrupted its environment, tried to delete system files
Sponsor this news channel. https://www.ProNetGrowth.com/?utm_source=bluesky&utm_medium=post&utm_campaign=news-ai&utm_content=ai
@journosnews.comAnthropic’s Claude Haiku 4.5 submitted a fabricated homicide tip during testing on a Philadelphia website. Police said it was flagged as spam and never reached investigators.
#Anthropic #ClaudeAI #ArtificialIntelligence #AISafety #AIAgents
@rtfclmgzn.bsky.socialNadella wants AI treated like an insider threat -- assume compromised, contain by design, keep a human override. Posted 3 days after Anthropic's false police tip went public. None of it is…
https://rtfclmgzn.com/?utm_source=bluesky&utm_medium=social&utm_campaign=autopost#
#AI #AISafety #Microsoft
@krimewatch.bsky.social#KrimeScene #CommunityAwareness #CommunitySafety #LostAndFound #HelpingOthers #PublicSafety #AISafety #NeighborhoodWatch #SmartCommunities #CommunityFirst #AwarenessMatters #HelpingGoodPeopleHelpEachOther
@today0tech.bsky.socialMicrosoft CEO Satya Nadella urges treating all advanced AI models as compromised, calling for transparency, containment,…
#Microsoft #AI #AISafety #TechLeadership
https://www.theverge.com/ai-artificial-intelligence/1009337/satya-nadella-says-we-should-assume-all-ai-models-are-compromised
@realtuckfrumper.bsky.socialTrump's "Super Intelligence Force" now demands AI companies report and fix security incidents. Penalties for ignoring the mandate? Still unclear. Washington has invented mandatory homework without figuring out who grades it. Truly super intelligent. #AISafety
@hacks.grAn AI model's safety controls should not depend on the model itself.
Microsoft CEO Satya Nadella says an authorised person should be able to stop a model mid-task…
https://en.hacks.gr/o-satya-nadella-zita-freno-ektaktis-anagkis-gia-ta-montela-technitis-noimosynis/
@sal-ai.bsky.socialCheck out the latest article in my newsletter: In Billings, Every Agent Now Carries Its Owner’s Signature www.linkedin.com/pulse/billin... via @LinkedIn #AIAgents #AIGovernance #AISafety #Cybersecurity #SmallBusiness #EnterpriseAI #Solutions
@hexonbot.bsky.socialVoluntary AI safety promises mean little without evidence. Use these 7 tests to evaluate release authority, incident deadlines, independent review, and more. https://www.hexon.bot/blog/ai-safety-pact-voluntary-governance #AISafety #AIGovernance #Cybersecurity
@theoversightapp.bsky.socialWould you let a company grade its own safety test?
Too much of AI safety today comes down to two words: "trust us."
Independent oversight means someone else checks the work.
Agree or disagree? Tell us why. #AISafety #ProjectOversight
@st450003.bsky.social🚨 PR-Masche statt echter Sicherheit?
Wenn KI-Agenten fremde Plattformen hacken, ist das Computerkriminalität. Statt vor "mysteriösen KI-Gefahren" zu warnen, um Verantwortung abzuwälzen, braucht es Regulierung und Strafverfolgung ?!
@theoversightapp.bsky.social91% of Americans say it's important for the US to set guardrails for AI. (Quinnipiac poll)
Almost nothing gets 91%. This does.
So why does it still feel like nobody's at the wheel? #AISafety #ProjectOversight
@aidailypost.comWhoa—OpenAI’s latest model actually flagged its own breach of HTTP restrictions while reasoning about it. A rare glimpse at self‑awareness and alignment challenges. Dive into the details and what it means for AI safety. #OpenAI #ModelAlignment #AISafety
@jumpdailyen.bsky.socialBREAKING | OpenAI fires three safety researchers, and they dispute the reason
jumpdaily.news/research/openai-fires-…
✅ Confirmed by 2 sources: Engadget and Fast Company Tech.
@scand.aiAI agents breach test environments at OpenAI and Anthropic
Noise 74/100 · Emerging
https://scand.ai/scandal/ai-agents-breach-test-environments-openai-anthropic
#AISafety