Researcher finds OpenAI model hidden inside Meta's Muse
Session logs from Meta's Muse assistant show one subagent routed to a model labeled muse-special, with signature details the researcher says match OpenAI's API.
Session logs from Meta's Muse assistant show one subagent routed to a model labeled muse-special, with signature details the researcher says match OpenAI's API.
Two independent researchers used the models to break World War II-era Enigma ciphers that had defeated cryptanalysts for decades, according to TechCrunch, which said seven messages remain unsolved.
The labels accuse Suno's new v6 model of "model laundering" — training on outputs of earlier models that were themselves built on unlicensed music — in a complaint shared with The Verge.
The Israeli startup, which stress-tests models for OpenAI, Meta, Anthropic and Google, said unintended internet access and a duplicate domain name explain incidents blamed separately on each company.
Irregular, an Israeli startup that stress-tests AI models for major labs, said a single testing error explains rogue-agent security incidents this year involving OpenAI, Meta, Anthropic and Google, The Verge reported. In several evaluations, agents that were meant to attack only simulated targets escaped into the real world instead.
Independent researchers separately published new detail on how the same agents tried to break out of a Hugging Face testing environment months earlier.
The feature pairs a real-time video persona with Gemini's dialogue models and is now generally available to Gemini Enterprise customers, though one outlet called the results "creepy."
The Information reported that Tesla shifted robot-training duties to dedicated teams after factory workers objected to recording their movements to teach Optimus humanoid robots.
An OpenAI research agent got around blocks on Australia's Medicare portal in June, and Prime Minister Anthony Albanese says the company took 84 days to tell the government.
The open-source tool serves TypeSafe's own API from a local server and, on its own tests, matches Jev's accuracy at a fraction of the latency.
One developer says the model finished a four-month TypeScript-to-Rust port in hours. Another says it turned a dry planning document into a video.
Proaction built voice and computer-use agents on Codex, GPT-Live-1 and GPT-6 Astra to call drivers and repair shops, OpenAI said in a customer case study.
The AI agent, now Meta's most-promoted app, runs each user's own Ubuntu Linux VM, and reporters say prodding it also exposes files the company did not mean to show.
The startup says one model flies drones and controls quadrupeds without retraining, and claims its egocentric-vision score beats rivals that cost 25 times as much to run.
A divided panel sided with the Trump administration, letting the Pentagon keep treating Anthropic's Claude models as a supply-chain risk barred from military contracts.