Desk
Broad artificial intelligence coverage.
White Circle says off-the-shelf tools broke image safeguards on leading OpenAI and Google models about 13% of the time—and newer releases look worse, not better.
Tech Against Terrorism’s new benchmark finds most models fail obvious terror prompts—and open-weight “abliterated” copies appear online in under three days.
Liquid AI’s open-weight LFM2.5-24B-A2B keeps a 24B MoE on disk but activates ~2B parameters per token—built for local agent loops, not chat theater.
Third Moment Research runs 10,000 agents for days on quantitative experiments — and uses Daybreak to pen-test the sandbox before they start.
Block's security team says Daybreak and related tooling cut detection tuning to under an hour and proposed patches to 18 minutes — defenders fighting fire with fire.
NVIDIA’s UK tour is less about one model and more about a full stack landing in hospitals, banks, cities, and climate prizes—with a $250 million compute pledge attached.
From Liverpool to Côte-d’Or to Milan, NVIDIA is wiring city operations to Cosmos, Nemotron, and local AI factories—useful if the pilots survive procurement and privacy reality.
Argon’s developer headline is extreme output length—up to 1M tokens in one response—at intro prices that undercut several frontier rivals until November 30.
Google DeepMind and Singapore’s GovTech fine-tuned an on-device Gemma model for local scam patterns—privacy-preserving detection that stays on the phone.
OpenAI’s EU data says Codex and ChatGPT Work already drive more than half of enterprise output tokens—while typical European firms still trail the global frontier.
Prosecutors say an OpenAI engineer used chatbots and internal tools to plan planting a bug in an executive conference room. Motive is still murky; the insider threat is not.
Meta’s Muse Wonder puts a frontier agent model on a public Model API at $4/$20 per million tokens, with Wonder Full gated behind paid Muse Code plans.