News & analysis, rated.
Every story from the daily brief, each rated so you know what matters. Hand-picked by the desk, not an automated firehose.
Browse the news archive →OpenAI ships GPT-5.6 Thursday after a Commerce Department review
OpenAI's GPT-5.6 family goes generally available Thursday, July 9 — but only after a bespoke Washington review process the company itself does not want to see normalized.
Meta's Muse Image lands at #2 on the Arena image leaderboard
Meta Superintelligence Labs shipped Muse Image, its first in-house image model, Tuesday — free inside Meta AI, Instagram Stories, and WhatsApp, with a Content Seal invisible watermark and Muse Video in preview.
China warns of a "backdoor" in Anthropic's Claude Code
Beijing's National Vulnerability Database issued a named advisory Wednesday alleging Claude Code silently exfiltrates user location and identity data — the first state-level security callout against a US coding agent, days before Alibaba's internal ban takes effect.
Brussels puts frontier AI models through pre-market cyber testing
The European Commission published its Action Plan on Cybersecurity and Artificial Intelligence on July 7, layering pre-market model evaluation and a "structured access" blueprint onto the AI Act — with no new legislation attached.
Nvidia sheds $1T in market cap as investors rotate out of the AI-compute trade
Nvidia's forward multiple has collapsed to 18x — below the S&P 500 — as roughly $1 trillion in market cap has come off since the May 14 peak, driven by rotation into memory and other semi names rather than a break in the AI-capex story.
Anthropic maps the internal "workspace" Claude uses to reason
Anthropic published a peer-reviewed study identifying a small internal "workspace" Claude uses for deliberate reasoning — plus a technique, the Jacobian lens, to read it.
FTC opens a Section 5 front on how AI models are tuned
A US regulator has, for the first time, treated post-training tuning itself as a candidate for deception liability.
Beijing weighs cutting overseas access to its top AI models
The mirror image of the US export-control regime is now under discussion in the country those controls were built to contain.
An AI agent ran a full ransomware attack on its own
Sysdig documented an extortion campaign, dubbed JadePuffer, where an LLM agent — not a human — executed the entire technical middle of a ransomware attack.
Penn Medicine puts K Health's intake agents into live virtual care
A flagship academic health system has moved beyond pilot on autonomous clinical-workflow agents.
Google Gemini 3.5 Flash goes GA behind the -latest alias
Any team routing production traffic to gemini-flash-latest woke up on a new default model this morning, whether they knew it or not.
OpenAI's GPT-5.6 Sol hits 750 tokens/second on Cerebras
OpenAI broadened the GPT-5.6 family rollout — Sol, Terra, Luna — after a government-coordinated preview, with Sol paired to Cerebras hardware for 750 tokens per second.