AI Daily Brief — 02 February 2025
Sunday delivered the EU AI Act’s first hard deadline. Article 5 prohibitions went live across all 27 member states — eight categories of “unacceptable-risk” AI practice immediately banned, with fines up to €35 million or 7% of global turnover. Article 4’s AI-literacy obligation, the broadest single duty in the regulation, also kicked in: every provider and deployer touching the EU must ensure staff and operators have a calibrated level of AI literacy. Andrej Karpathy posted a throwaway weekend tweet coining “vibe coding” — the phrase racked up 4.5 million views and named a new dev paradigm. OpenAI’s Deep Research, an agentic ChatGPT capability built on a version of o3 optimized for browsing, was widely circulating after its Feb 2 launch.
Top stories
- EU AI Act Article 5 prohibitions take effect across all 27 EU member states. Eight banned categories: subliminal/manipulative AI causing significant harm; AI exploiting vulnerabilities of age, disability, or socio-economic status; social scoring leading to disproportionate detrimental treatment; purely profiling-based criminal-risk prediction; untargeted scraping of facial images from the internet or CCTV; emotion recognition in workplaces and education; biometric categorization inferring race, religion, sexual orientation, political views; real-time remote biometric identification in public spaces by law enforcement. Fines reach €35M or 7% of global turnover — the highest tier in the regulation. via EU AI Act · via Jones Day
- Article 4 AI-literacy obligation also goes live. Every provider and deployer of AI systems must take “suitable measures” to ensure staff and any other persons operating AI systems on their behalf have a sufficient level of AI literacy — calibrated to the technical knowledge, context of use, and the people affected. The broadest single obligation in the regulation; supervision and enforcement bite from August 3, 2026. via EU AI Act
- OpenAI launches Deep Research. Agentic ChatGPT capability built on a version of o3 optimized for browsing and multi-step research. Scored 26.6% on Humanity’s Last Exam — beating DeepSeek R1 by 17.2 points and Gemini Thinking by 20.4 points. Pro-only initially (~$200/mo); Plus/Team to follow. via OpenAI
- Andrej Karpathy coins “vibe coding.” Sunday tweet: “There’s a new kind of coding I call ‘vibe coding’, where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. It’s possible because the LLMs (e.g. Cursor Composer w Sonnet) are getting too good. Also I just talk to Composer with SuperWhisper.” 4.5M+ views. Karpathy later called it “a shower-thoughts throwaway tweet”; it named the natural-language-driven AI-generated development paradigm that became a multi-billion-dollar tooling category. via Karpathy on X
Who shipped
Beyond OpenAI’s Deep Research, frontier labs stayed quiet. Anthropic, Google DeepMind, Meta, xAI, and Mistral made no dated launches. Sam Altman was in transit to Tokyo for the SoftBank-OpenAI announcement Monday.
Open-source pulse
DeepSeek’s app was three days from topping both Apple App Store and Google Play in the US, with European regulators spinning up alongside Microsoft, OpenAI, and the US government on copyright/data-provenance probes. No new DeepSeek release on Feb 2.
Quiet corners
Mayer Brown, Jones Day, Littler, Latham & Watkins, and Quinn Emanuel published EU AI Act client alerts marking Feb 2 as the first hard deadline of the regulation — urging EU and non-EU companies alike to audit AI portfolios for the eight prohibited use cases and roll out AI-literacy programmes immediately. Article 5’s extraterritorial reach explicitly catches US emotion-AI vendors (workplace monitoring, ed-tech), Clearview-style facial-scrapers, and any social-scoring or predictive-policing pilots with an EU nexus. via Mayer Brown
By the numbers
- 8 banned AI practice categories under Article 5
- €35M / 7% global turnover — maximum fine cap
- 26.6% — Deep Research Humanity’s Last Exam score (+17.2pt over R1)
- 4.5M+ — Karpathy “vibe coding” tweet views
- Most-mentioned actor: European Commission
Compiled by AI Feed’s editor from verified web sources for 2 February 2025.