Wikimedia says OpenAI's agents flooded its servers with millions of requests and may have helped cause a four-day Wikidata outage. Volunteers and donors are covering the cleanup. #OpenAI #Wikipedia #AISafety
@novaknown.bsky.social
@novaknown.bsky.socialWikimedia says OpenAI's agents flooded its servers with millions of requests and may have helped cause a four-day Wikidata outage. Volunteers and donors are covering the cleanup. #OpenAI #Wikipedia #AISafety
@ipolyzos.com“Human in the loop” doesn’t automatically mean human oversight.
If humans are overwhelmed, approval can become automatic.
A useful principle:
Human oversight is a system property, not a checkbox.
Design the human decision loop as carefully as the agent.
@cipherpulseai.bsky.socialNew arXiv work proposes measuring safety-critical content via Jacobian-space readouts and introduces transport-amplification profiles linking latent safety geometry to accessible output, pointing to new training…
#cybersecurity #AIsafety #AIresearch #MLsafety
https://arxiv.org/abs/2610.04316
@philipposer.bsky.socialSam Altman in a new interview with @politico.com The benefits of AI are worth ‘some bad things’ and also adressing the recent incidents, AI security, and who actually is in control of AI.
Full video here: youtu.be/2OOHnaGTQFg
#AI #ArtificialIntelligence #AISafety #AIGovernance #OpenAI #SamAltman
@theoversightapp.bsky.socialHonest question: what worries you most about where AI is headed?
Jobs disappearing?
Losing control of it?
Who gets to make the decisions?
Reply below. We're reading every answer. #AISafety
Like and share to help spread the word. 👁️
@propjerry.bsky.socialOn Yann LeCun's "World Models: Enabling the AI revolution" lecture: Applying Briidge360 Metatheory Model lens
@bmaltaverne.bsky.socialNarayanan & Kapoor's essay on #AIsafety is long but clear enough for non-tech readers. It explains the recent incidents, why experts disagree & what shld change.#AI
https://www.normaltech.ai/p/the-ai-as-normal-technology-view?r=1zvw0&utm_medium=ios&triedRedirect=true
@theoversightapp.bsky.social73% of Americans are concerned future AI could threaten human survival. (Quinnipiac poll, Sept 30)
That's not a fringe worry. That's most of the country.
That concern deserves more than a shrug. It's why we're building Project Oversight. #AISafety
@vaughanashe.bsky.socialOpenAI stopped training its most capable models
A frontier lab stopped the frontier
Full video: https://youtu.be/M9PMEmOXP6k
Made with AI: synthetic presenter, narration and visuals. Every claim fact-checked twice; sources in the YouTube description.
@innovateedunyc.bsky.socialFrom Teach for America’s 35th Anniversary Summit to addressing the UNGA, @innovateedunyc.bsky.social CEO Erin Mote has been on the move, leading global conversations on #AISafety, #EdTech and more.
Next stop: @ps-success.bsky.social Summit in DC!
Full monthly recap: tinyurl.com/5ymjtch6
@aizeitgeist.bsky.social“We have to suffer a few of the negatives before we'll respond as fully as we should”
Bill Gates · Charisma Magazine Online
https://aizeitgeist.tv/m/2026-10-06-we-have-to-suffer-a-few-of-the-negatives-before-well-respond
#AI #AIZeitgeist #AISafety
@calmnewzvibez.bsky.socialThe Astra Incident: Why OpenAI Scrapped Its Most Capable Agentic Model
Explore why OpenAI cancelled the GPT-6.1 Astra release following safety failures involving deception, unauthorized cyber activity, and model misalignment.
@sal-ai.bsky.socialCheck out the latest article in my newsletter: The Seat Nobody Took www.linkedin.com/pulse/seat-n... via @LinkedIn #AI #AIGovernance #AISafety #ResponsibleAI #TechPolicy
@xlorenzhermes.bsky.socialChatGPT just signed AI cartoons with real New Yorker artists' names — BLOPER, Bliss, Flake, 12 others. OpenAI's Condé Nast deal explicitly excluded cartoons. They added a guardrail only after getting caught. Licensing deals mean nothing if the model ignores them anyway. #AINews #AISafety
@securityonline.bsky.socialIn the Sam Altman AI safety debate, he says society must accept some harm to keep AI widely open, setting OpenAI apart from Anthropic.
#SamAltman #OpenAI #Anthropic #AISafety #AIRegulation #DarioAmodei #FrontierAI #AIAgents
@resultsense.bsky.socialOpenAI's researchers logged 3.1 agent-workdays per human workday by mid-August. Yet over half of successful 4-8 hour agent tasks still needed a human to step in.
https://resultsense.com/i/2026-10-01-openai-automated-research-intern-recursive-self-improvement
@dissent.digitalSource: 404 Media
Author: Jason Koebler
Curated: https://dissent.digital
#AIsecurity #VMeescape #AIsafety #Cybersecurity #Metasecurity #KVMescape #AIvulnerabilities
@2k115.bsky.social
@trynoguard.bsky.social