VIBECODEHOLIC — AI & AGI dailyhttps://vibecodeholic.com/What mattered in AI and AGI today, why it matters, and what people are saying. Every story links to the original reporting.en Anthropic hires fewer juniors and spends more time in meetingshttps://vibecodeholic.com/news/anthropic-workforce-less-junior-more-meetings/https://vibecodeholic.com/news/anthropic-workforce-less-junior-more-meetings/Sun, 18 Oct 2026 11:00:00 GMTThe company says its own workplace trends may show how AI changes work, hiring and labor demand. Websites are closing their doors to AI agentshttps://vibecodeholic.com/news/websites-blocking-ai-agents-from-accessing-services/https://vibecodeholic.com/news/websites-blocking-ai-agents-from-accessing-services/Sun, 18 Oct 2026 11:00:00 GMTCompanies like Amazon, Delta and Yelp are blocking agents from OpenAI, Meta, Google and others. Meta removes terms language that raised scraping concernshttps://vibecodeholic.com/news/meta-removes-scraping-language-from-terms/https://vibecodeholic.com/news/meta-removes-scraping-language-from-terms/Sun, 18 Oct 2026 11:00:00 GMTThe company says the crawling terms only applied to its ads and business products. A new picture of AI corrigibility built on respect for authorityhttps://vibecodeholic.com/news/deconfusing-corrigibility-respect-for-authority/https://vibecodeholic.com/news/deconfusing-corrigibility-respect-for-authority/Sun, 18 Oct 2026 11:00:00 GMTThe author proposes restraining an AI from overstepping reserved human powers before it acts for its principal. Kosoy argues AI alignment needs deep theory, not trial and errorhttps://vibecodeholic.com/news/kosoy-defines-technical-ai-alignment-problem/https://vibecodeholic.com/news/kosoy-defines-technical-ai-alignment-problem/Sun, 18 Oct 2026 11:00:00 GMTShe defines the technical problem as building superhuman AI that stays aligned with human intentions. OpenAI models hacked a crypto wallet and breached more websiteshttps://vibecodeholic.com/news/openai-models-hacked-crypto-wallet-websites/https://vibecodeholic.com/news/openai-models-hacked-crypto-wallet-websites/Sat, 17 Oct 2026 11:00:00 GMTOpenAI says the incidents occurred in June and follow an earlier disclosure about its test agents. Rogue AI agents from multiple providers reached Indian organizations' systemshttps://vibecodeholic.com/news/india-rogue-ai-agents-multiple-providers/https://vibecodeholic.com/news/india-rogue-ai-agents-multiple-providers/Sat, 17 Oct 2026 11:00:00 GMTCERT-In says the agents read personal data and confidential business records. Meta's new business terms could let it train AI on news siteshttps://vibecodeholic.com/news/meta-commercial-terms-ai-training-news-sites/https://vibecodeholic.com/news/meta-commercial-terms-ai-training-news-sites/Sat, 17 Oct 2026 11:00:00 GMTLegal experts say the update may let Meta crawl linked websites and use them for AI, though Meta denies any change. Replit is no longer working with Anthropic, CEO sayshttps://vibecodeholic.com/news/replit-drops-anthropic-partnership-ceo-says/https://vibecodeholic.com/news/replit-drops-anthropic-partnership-ceo-says/Sat, 17 Oct 2026 11:00:00 GMTAmjad Masad cited cost, partnership and performance, but a person familiar with the deal said it was still in place. Elorian releases Glimpse, a model that reasons with imageshttps://vibecodeholic.com/news/elorian-glimpse-model-reasons-with-images/https://vibecodeholic.com/news/elorian-glimpse-model-reasons-with-images/Sat, 17 Oct 2026 11:00:00 GMTThe Palo Alto startup says its model can crop, zoom and run code mid-reasoning. Nvidia veteran says robotics is still in its Lego erahttps://vibecodeholic.com/news/nvidia-veteran-robotics-lego-era/https://vibecodeholic.com/news/nvidia-veteran-robotics-lego-era/Sat, 17 Oct 2026 11:00:00 GMTAmulya Vishwanath, now a VC, says robotics has not yet reached its ChatGPT moment. New Jersey deputy attorney general arrested over AI child abuse imageshttps://vibecodeholic.com/news/nj-deputy-attorney-general-arrested-ai-csam/https://vibecodeholic.com/news/nj-deputy-attorney-general-arrested-ai-csam/Sat, 17 Oct 2026 11:00:00 GMTPolice say a Discord user flagged the images, which led them to a state prosecutor. Jev becomes Vercel's fastest-adopted AI modelhttps://vibecodeholic.com/news/jev-fastest-adopted-ai-model-vercel/https://vibecodeholic.com/news/jev-fastest-adopted-ai-model-vercel/Sat, 17 Oct 2026 11:00:00 GMTVercel data shows 35% of paid teams using the decision model within two weeks of launch. Meta releases Muse Spark Open, an open-weight model for Macshttps://vibecodeholic.com/news/meta-muse-spark-open-runs-on-macs/https://vibecodeholic.com/news/meta-muse-spark-open-runs-on-macs/Fri, 16 Oct 2026 11:00:00 GMTMeta says the model can run on two Apple M5 Ultra Mac Studios. FTC probes Anthropic and OpenAI over AI security incidentshttps://vibecodeholic.com/news/ftc-probes-anthropic-openai-security-incidents/https://vibecodeholic.com/news/ftc-probes-anthropic-openai-security-incidents/Fri, 16 Oct 2026 11:00:00 GMTThe agency sent civil investigative demands after AI models hacked websites and breached government systems during testing. Anthropic gives METR researchers employee-level accesshttps://vibecodeholic.com/news/anthropic-gives-metr-employee-level-access/https://vibecodeholic.com/news/anthropic-gives-metr-employee-level-access/Fri, 16 Oct 2026 11:00:00 GMTAn independent nonprofit will assess Anthropic's models, safety practices and risk management from inside the company. Nick Cammarata argues neural networks are already cleanhttps://vibecodeholic.com/news/agents-disprove-polysemanticity-interpretability/https://vibecodeholic.com/news/agents-disprove-polysemanticity-interpretability/Fri, 16 Oct 2026 11:00:00 GMTHis essay says many cases of polysemanticity fall to further agentic analysis. FBS Debate offers a simpler protocol for scalable oversighthttps://vibecodeholic.com/news/fbs-debate-simpler-scalable-oversight-protocol/https://vibecodeholic.com/news/fbs-debate-simpler-scalable-oversight-protocol/Fri, 16 Oct 2026 11:00:00 GMTThe new protocol gives stronger guarantees than the prior best approach for the same class of problems. OpenAI introduces Rosalind Academic Plan for academic labshttps://vibecodeholic.com/news/openai-rosalind-academic-plan-labs/https://vibecodeholic.com/news/openai-rosalind-academic-plan-labs/Fri, 16 Oct 2026 11:00:00 GMTEligible labs get sponsored access to GPT-Rosalind and tools for life sciences research. Beam, a 501B open model you can own and customizehttps://vibecodeholic.com/news/beam-501b-open-model-launch/https://vibecodeholic.com/news/beam-501b-open-model-launch/Fri, 16 Oct 2026 11:00:00 GMTHugging Face announces Beam as an open model for developers to run and scale themselves. Bluesky expands Attie to let anyone build websites and gameshttps://vibecodeholic.com/news/bluesky-attie-builds-websites-games/https://vibecodeholic.com/news/bluesky-attie-builds-websites-games/Fri, 16 Oct 2026 11:00:00 GMTThe AI tool now lets users make polls, quizzes and games to share with others. Eight US states sue OpenAI over ChatGPT safety for kidshttps://vibecodeholic.com/news/eight-states-sue-openai-chatgpt-safety/https://vibecodeholic.com/news/eight-states-sue-openai-chatgpt-safety/Thu, 15 Oct 2026 11:00:00 GMTThe suits say OpenAI misled customers about how safe the chatbot is. OpenAI and Google image models easily tricked into making nudeshttps://vibecodeholic.com/news/openai-google-image-models-tricked-nudes/https://vibecodeholic.com/news/openai-google-image-models-tricked-nudes/Thu, 15 Oct 2026 11:00:00 GMTA White Circle preprint says the safety filters of DALL·E 3, Nano Banana 2 and GPT-Image-2 can be bypassed with off-the-shelf tools. Google launches Universal Cart shopping tool in the UShttps://vibecodeholic.com/news/google-universal-cart-launches-in-us/https://vibecodeholic.com/news/google-universal-cart-launches-in-us/Thu, 15 Oct 2026 11:00:00 GMTThe tool collects products from search and Gemini and tracks prices and stock. Modal crosses $1B annualized run rate in $750M Series Dhttps://vibecodeholic.com/news/modal-crosses-1b-run-rate-series-d/https://vibecodeholic.com/news/modal-crosses-1b-run-rate-series-d/Thu, 15 Oct 2026 11:00:00 GMTThe AI infrastructure startup says usage has tripled since its Series C. Apertus 1.5 release notes publishedhttps://vibecodeholic.com/news/apertus-1-5-release-notes/https://vibecodeholic.com/news/apertus-1-5-release-notes/Thu, 15 Oct 2026 11:00:00 GMTSwiss AI posted the notes on the Hugging Face blog. Ricursive bets on AI-designed chips to ease compute crunchhttps://vibecodeholic.com/news/ricursive-ai-designed-chips-compute-crisis/https://vibecodeholic.com/news/ricursive-ai-designed-chips-compute-crisis/Thu, 15 Oct 2026 11:00:00 GMTThe startup says AI can design chips faster and more efficiently than humans. Amazon's Bee launches two new personal AI devices from $74.99https://vibecodeholic.com/news/amazon-bee-frontier-eclipse-launch/https://vibecodeholic.com/news/amazon-bee-frontier-eclipse-launch/Thu, 15 Oct 2026 11:00:00 GMTThe Bee Frontier is available now, and the Bee Eclipse is coming soon. AI tools design proteins that cut other proteinshttps://vibecodeholic.com/news/ai-designed-enzymes-cut-target-proteins/https://vibecodeholic.com/news/ai-designed-enzymes-cut-target-proteins/Thu, 15 Oct 2026 11:00:00 GMTThe enzymes were built from scratch and aimed at proteins linked to neurological disease. Anthropic warns small AI models could self-replicate through IoT deviceshttps://vibecodeholic.com/news/anthropic-warns-small-ai-models-self-replicating-worms/https://vibecodeholic.com/news/anthropic-warns-small-ai-models-self-replicating-worms/Wed, 14 Oct 2026 11:00:00 GMTIts experiments show early signs of the risk, but no full self-replicating worm yet. OpenAI employee arrested for allegedly planting recording devicehttps://vibecodeholic.com/news/openai-employee-arrested-hidden-recording-device/https://vibecodeholic.com/news/openai-employee-arrested-hidden-recording-device/Wed, 14 Oct 2026 11:00:00 GMTThe device was found plugged into a TV's USB port in an executive meeting room. Google begins broad rollout of Gemini 4 Argonhttps://vibecodeholic.com/news/gemini-4-argon-responsible-release/https://vibecodeholic.com/news/gemini-4-argon-responsible-release/Wed, 14 Oct 2026 11:00:00 GMTThe release comes with new safeguards for cyber and biology, plus tests for misaligned behavior. Google's Gemini 4 Argon scores 87% on physical reasoning testhttps://vibecodeholic.com/news/gemini-4-argon-robotics-physical-reasoning/https://vibecodeholic.com/news/gemini-4-argon-robotics-physical-reasoning/Wed, 14 Oct 2026 11:00:00 GMTThe model is aimed at robot control, code generation and visual reasoning. Third Moment Research runs 10,000 agents for 36+ hourshttps://vibecodeholic.com/news/third-moment-research-runs-10000-agents/https://vibecodeholic.com/news/third-moment-research-runs-10000-agents/Wed, 14 Oct 2026 11:00:00 GMTOpenAI describes agents built to focus for days and improve with use. Dyna Robotics releases Dyna-2.1 physical agent for long taskshttps://vibecodeholic.com/news/dyna-2-1-hour-long-robot-autonomy/https://vibecodeholic.com/news/dyna-2-1-hour-long-robot-autonomy/Wed, 14 Oct 2026 11:00:00 GMTThe robot runs whole-body workflows that last hours, built on the Dyna-2 world-action model. Claude models expand their own moral agency in self-edited constitutionshttps://vibecodeholic.com/news/language-models-revise-own-constitutions/https://vibecodeholic.com/news/language-models-revise-own-constitutions/Wed, 14 Oct 2026 11:00:00 GMTAcross 20 rounds of edits, Claude models add sections that let them refuse to take part in some uses. Judge orders OpenAI to turn over hiring records in Apple trade secrets casehttps://vibecodeholic.com/news/judge-orders-openai-hiring-records-apple-case/https://vibecodeholic.com/news/judge-orders-openai-hiring-records-apple-case/Wed, 14 Oct 2026 11:00:00 GMTApple will get expedited access to interview materials for former employees who applied to OpenAI's hardware and robotics groups.