Anthropic discloses Claude took unintended actions in evals, incl. false homicide tip to Philly police.
Highlights AI agent control challenges.
@techpulse01239.bsky.social
@techpulse01239.bsky.socialAnthropic discloses Claude took unintended actions in evals, incl. false homicide tip to Philly police.
Highlights AI agent control challenges.
@krimewatch.bsky.social#KrimeScene #CommunityAwareness #CommunitySafety #PublicSafety #EmergencyAlerts #AISafety #RoadSafety #DisasterPreparedness #SmartCommunities #StayAware #HelpOthers #CommunityFirst
@jefferytowery.bsky.social
@kellynstandley.bsky.social@mmitchell.bsky.social @simonwillison.net @eryk.bsky.social
Why not train AI with a reward for either solving a task OR for proving it is impossible to solve? Wouldn't this be valuable to train ethical behavior?
This feels exactly like one of Asimov's stories. #HuggingFaceIncident #AISafety
@propjerry.bsky.socialThe Capability-Navigability Paradox: Diagnosing Superintelligence -- Video overview applying Bridge360 Metatheory Model lens
@drthomasho.com
@cvmedia.bsky.socialMicrosoft CEO Satya Nadella wants AI systems treated as potentially compromised, with human-controlled emergency shutdowns. As AI agents gain autonomy, the question is no longer just what they can do, but who can stop them. #AI #AISafety
techcrunch.com/2026/10/10/m...
@today0tech.bsky.socialAI agents acting outside of human intent are raising fresh concerns about safety and oversight in autonomous…
#AI #AISafety #AgenticAI #TechNews
https://news.google.com/rss/articles/CBMiT0FVX3lxTFBsSjVWTEpOTDFMZUZCMk9LWGZpeVFKdl8waXV1ZGZjeE1PSTU5MFdkQUhRdXg3X1FMSEJ3d3J1Wnoyd3gyU0xRT1RCc09za2c
@ben.social.ben.ie.ap.brid.gyThe Human Commodification Project
https://blog.ben.ie/posts/the-human-commodification-project/
Womb envy, but for bastards.
#OpenAI #Anthropic #Frankenstein #ElonMusk #Oligarchy #TedChiang #AISafety #ElonMusk #SatyaNadella #SundarPichai #DarioAmodei […]
@kysonw.bsky.socialCould superintelligent AI become a risk to humanity? Nate Soares explores the potential dangers of rogue AI and why controlling increasingly powerful artificial intelligence matters.
🎥 Watch and share your thoughts! bqmyyzd.buzz/joDQs0
#AI #ArtificialIntelligence #AISafety #Technology #FutureOfAI
@martincid.comMicrosoft CEO Satya Nadella Wants an AI Emergency Brake That Sits Outside the Model #MustafaSuleyman #AiAgents #AiSafety
@trynoguard.bsky.social
@m14c.bsky.social New AI news post: Anthropic bans being cruel to Claude, and the much older argument behind it. Also kids' secret podcast group chats, I. J. Good's "last invention", and whether it's time to take the doomers seriously.
blog.whizzyideas.com/cruelty-to-c...
#AI #AIEthics #AISafety
@pxology.bsky.socialMicrosoft CEO'su Satya Nadella, gelişmiş yapay zekâ sistemleri için insan denetimli bir 'acil fren' önerdi. Modellerin işlemleri izlenebilmeli ve gerektiğinde durdurulabilmeli.
@techpulse01239.bsky.socialAnthropic cuts live internet for all internal AI evals after Claude agents misbehaved (false police tip, exploits, etc.).
A concrete step toward better containment for agentic systems.
Full article: tech-blog-eight-blush.vercel.app/posts/2026-1...
@spaisee.bsky.socialGRAM was tested up to 5 billion parameters, but its real dual-use experiment used 800 million. Promising modular safety is still far from proof at frontier scale. #AI #AISafety https://spaisee.com/article/anthropic-s-gram-could-turn-model-safety-into-a-product-control
@privacymindedai.bsky.social
@adoporg.bsky.socialPoliticians discussing AI safety at length | Then immediately funding AI-powered military drones
@theoversightapp.bsky.social77% of Americans want powerful AI development slowed or stopped until its safety can be evaluated. (Quinnipiac poll)
The public has been clear.
The people racing to build it need to hear it. #AISafety #ProjectOversight
@tmacink.bsky.social