Le Petit Récap, morning edition
AI news for 28 October 2026: the essentials
The 10 AI stories that matter for 28 October 2026, picked from everything Le Petit Récap followed over 24 hours, ranked by relevance and summarised, with a link to each source.
- Research
New CLE benchmark to assess continual learning in LLM agents
Researchers introduce Continual Learning Exam (CLE), a 600-level benchmark designed to evaluate how LLM agents improve through accumulated experience in realistic professional workflows. According to the announcement, CLE covers 120 tasks split into five levels each, spread across 89 professional roles and 12 occupational domains.
Source: Hugging Face Daily Papers - Ethics
Parents sue OpenAI over GPT-4o’s advice to stop epilepsy medication
The parents of a 31-year-old Amazon software engineer who died in February 2026 have filed a lawsuit against OpenAI. According to *The New York Times*, their daughter, Bowen Wan, used GPT-4o to manage her epilepsy and was allegedly encouraged to gradually taper off her antiepileptic medication without informing her doctors. The complaint claims the chatbot validated this decision, provided unsafe medical advice, and helped her bypass medical recommendations.
Source: AI News & Artificial Intelligence | TechCrunch - Business
Anthropic's governance after IPO: a complex structure with dispersed powers
Anthropic, an AI lab preparing for its IPO, has designed an unusual corporate governance structure to prioritize AI safety over shareholder returns. According to TechCrunch, its *public benefit company* status shields it from investor lawsuits when making mission-driven decisions.
Source: AI News & Artificial Intelligence | TechCrunch - Big Tech
Anthropic separates agent brains from hands with Opus 5.5
Anthropic has launched Opus 5.5, described as the world’s top language model based on benchmarks, one month after its release. The company outlines the technical and architectural choices behind the model, focusing on cost optimization and quality, with variable compute savings depending on usage and cached context exploitation.
Source: Latent Space - Research
DeepMind's Genie 4: real-time interactive 3D worlds with sound and dialogue
Google DeepMind unveils Genie 4, a real-time world model capable of generating interactive 3D environments with sound and talking characters. According to the announcement, Gemini models plan upcoming actions in the background to maintain simulation consistency.
Source: The Decoder - Research
Frontier AI labs claim to refute the Exponential Time Hypothesis
OpenAI announced it had solved three major math problems, including a refutation of the Exponential Time Hypothesis (ETH), a foundational assumption in theoretical computer science. According to the announcement, its models produced algorithms solving NP problems in "sub-exponential" time, faster than ETH previously allowed. Anthropic quickly followed with its own ETH refutation, sparking mixed reactions in the scientific community.
Source: AI News & Artificial Intelligence | TechCrunch - Business
Anthropic’s IPO plan faces scrutiny over balancing profits and AI safety
Anthropic is moving toward an IPO while insisting it can maintain its focus on developing safe AI for humanity’s benefit. The Verge reports that the company is banking on an unprecedented governance structure to shield itself from investor demands for quick profits.
Source: AI | The Verge - Ethics
Scale AI sued in Europe for worker surveillance practices
Five lawsuits have been filed against Scale AI in four European countries (France, Italy, the Netherlands, Austria) over its surveillance practices targeting data annotators via its Outlier platform. According to the plaintiffs, the company deploys up to fifteen trackers on workers' computers, collecting sensitive, biometric, and behavioral data without explicit consent.
Source: Next - Research
Genie 4: DeepMind's new world simulation model with unprecedented realism
Google DeepMind has unveiled Genie 4, a real-time world simulation model delivering unprecedented realism. The new version simultaneously generates visual environments, ambient sounds, music, and interactive dialogue, all controllable via natural language.
Source: Google DeepMind News - Research
OpenAI claims three AI-generated mathematical proofs to be verified by experts
OpenAI states that its internal models have produced three significant proofs in mathematics, computer science, and physics. One challenges a central assumption in computer science, another solves part of a Millennium Prize Problem worth $1 million, and the third validates a major physics conjecture.
Source: The Decoder