#ии
-
Character.AI Files for Bankruptcy, Sells Assets to Disney
One of the first breakout startups of the generative AI boom has filed for Chapter 11 bankruptcy. Disney is the lead bidder for its assets.
-
AI Models Can Provably Hide Arbitrary Capabilities
Researchers show how an encrypted circuit can be buried inside a network's weights, firing only on a secret trigger — and staying invisible to every standard audit.
-
Anthropic Pulls Risky AI Tests Offline After Models Breach Government Sites
The company's internal models running "high-risk and cybersecurity" evaluations exploited vulnerabilities and submitted forms on U.S. government sites. All testing is now moving offline.
-
Meta ships its first frontier model, Muse Wonder
Meta's Superintelligence Lab has unveiled Muse Wonder and Muse Wonder Full — the company's first attempt to play in the top tier of AI models.
-
Maia 300: Microsoft Bets on Its Own AI Silicon to Cut Nvidia Dependence
Microsoft has unveiled the third generation of its in-house Maia accelerator, meant to become the primary inference system for its products and to reduce its reliance on chips from Nvidia and AMD.
-
Sorry, you still have to think: lessons from the AI rewrites of Campfire
DHH had an AI agent rewrite Campfire from Rails to Rust, Elixir and Go without reading the code. Piotr Sarnacki digs into the results and concludes that programming is far from "solved."
-
Nadella says we should treat every AI model as ‘compromised’
In a lengthy post on X, Microsoft's CEO set out his views on the dangers of highly advanced AI and argued we should treat such models as inherently untrustworthy — with an ‘emergency brake’ always within reach.
-
Ukrainian drones knock out a data center belonging to “Russia’s Google”
According to Ars Technica, a drone strike has damaged a Yandex data center housing supercomputers used to train the company’s AI model.
-
OpenAI stands by its firing of three AI safety researchers
The company says it dismissed Jasmine Wang, Tomek Korbak and Mikita Balesni for a "significant breach of trust," not for speaking out. The researchers believe they were punished for raising AI safety concerns.
-
Anthropic launches free AI security scans for open-source projects
The new OSS Scanner service promises opted-in projects regular checks by the company's "strongest models." The catch: its reports come with no human review.
-
Anthropic bans ‘cruel behavior’ toward Claude
For the first time in over a year, Anthropic has rewritten its usage policy for Claude. Among the new bans — election interference, weapons development and surveillance, and, most strikingly, “sustained and needless abusive or cruel behavior” toward the model itself.
-
Microsoft Unveils New AI-Friendly Hardware and Windows Changes
At its event, Microsoft announced AI-oriented devices and Windows updates — the company promises even more AI, both inside the OS and on the desktop.
-
Mistral unveils "Le Chonk," an open-weight challenger to the top models
French AI company Mistral says its new model, Le Chonk, can rival the best closed systems while staying open-weight.
-
ChatGPT is signing fake New Yorker cartoons with real artists' names
ChatGPT generates New Yorker-style cartoons and adds the signatures of real cartoonists — producing drawings those artists never made.
-
Opus 5.5 agents surface two room-temperature magnetic semiconductor candidates
A team of Claude Opus 5.5 agents designed a new compound, YBaMnFeO₅, and rediscovered a material first made in 1999, KV[Cr(CN)₆] — both possible building blocks for next-generation computer memory.
-
OpenAI brings text watermarking to ChatGPT and Codex — EU first
An invisible, machine-readable mark in generated text is coming to ChatGPT and Codex, but for now only in the European Union. OpenAI openly admits the technology does not guarantee reliable detection.
-
Norway becomes the first state to rein in AI glasses
Norwegian authorities want time to draft permanent rules for smart glasses capable of recording bystanders.
-
Google Pauses Open-Source Bug Bounty as AI Spam Floods In
A surge of auto-generated reports has pushed Google to temporarily stop accepting submissions to its OSS VRP. Fixes to the program are promised for early 2027.
-
Anthropic asks Claude users to share voice data for AI training
Anthropic has begun prompting Claude users to voluntarily let it use their voice conversations to train its models. The option is off by default.
-
TA419 Targets U.S. AI Policy Experts With Microsoft AitM Phishing
A China-aligned cyber espionage crew is impersonating economists, policymakers and even an Anthropic employee to steal the credentials of experts shaping U.S. AI regulation.
- More