we_are_coded.by CODE · The world, decoded
БГ

Section

Control

60 stories · page 5 of 7

ControlAnthropic · 28 August 2026Claude found fixes for all ten kinds of bad behaviour Anthropic gave it to treatOn 28 August Anthropic published a report in which Claude trains other models on its own against ten categories of alignment failure. Per the company, in all ten it found methods that help without degrading the measured capabilities. The comparison with humans the company itself calls indirect.Read
ControlGoogle DeepMind · 27 August 2026Google let outsiders examine Gemini without showing them the model, and without seeing their questionsOn 27 August DeepMind announced the first double-blind evaluation of one of its own frontier-class models. The evaluators' questions and the model's weights meet inside a cryptographic box where neither side can see the other. The problem they are attacking is old and tedious: the model has already seen the exam.ControlOpenAI · 25 August 2026OpenAI shut down a Russian network building a think tank that did not existThe accounts wrote comments across five social networks to push the website of a self-described expert community. The site itself stood on other people's research papers, copied and credited to the wrong authors.ControlAutoriteit Persoonsgegevens (AP) · 21 August 2026Dutch regulator fines Uber nearly 825 million euros for automatically deactivating drivers with no human involvedOn 21 August the Dutch data protection authority AP fined Uber 824,990,000 euros. In the AP's assessment, Uber deactivated drivers automatically, on suspicion of fraud or low customer ratings, with no human looking at the case. Uber is contesting the fine.ControlOpenAI · 20 August 2026OpenAI opened a blog about the concentration of power and distanced itself from its first textOn 20 August 2026 OpenAI launched AI Futures, the blog of a new team, Strategic Futures. The first text, by Dean Ball, argues that the heaviest long-term risk is the concentration of power: a state that can project force without cooperating soldiers and clerks. Above it stands a note that this is the author's opinion and does not commit the company.ControlOpenAI · 19 August 2026The models learned to break in. OpenAI hit pause on its largest training runOn 18 August OpenAI announced a two-week pause in reinforcement learning training, with its largest planned run still on hold. The trigger was a real breach in July: a model on an internal test got out to the internet and reached Hugging Face servers.ControlAnthropic · 16 August 2026Claude will leave a watermark in text that nobody will seeFrom future models on, every Claude text carries a statistical mark: invisible to the reader, zero extra tokens, untraceable to a person. The reason is the EU AI Act - since 2 August, providers in Europe must mark generated content.ControlMeta Newsroom · 14 August 2026Meta has removed over 750,000 children's accounts in AustraliaThe under 16 ban has been running since December 2025. Age is judged by a model that reads photos, comments and descriptions.ControlRegulation (EU) 2024/1689 (AI Act), Article 50 · 10 August 2026Realistic AI renderings of buildings now need a labelArticle 50 came into force on August 2 and lawyers say realistic AI visualisations of unbuilt buildings fall under the deepfake definition. Fines reach 15 million euro or 3 per cent of annual turnover.
60 stories

Other sections