arstechnica.com / .arstechnica-com-technology

Risks to BIG-IP users include supply-chain attacks, credential loss, and vulnerability exploits.

Tiny, fast model hits coding scores similar to GPT-5 and Sonnet 4.

Sam Altman claims new tools can detect mental distress while relaxing limits for adults.

Scams like this one net billions from well-educated victims.

The 1 petaflop DGX Spark system runs AI models with 200 billion parameters locally for $4K.

New paper reveals reducing "bias" means making ChatGPT stop mirroring users' political language.

Malicious app required to make "Pixnapping" attack work requires no permissions.

New design sets a high standard for post-quantum readiness.

Among other things, the scammers bypass multi-factor authentication.

Anthropic study suggests "poison" training attacks don't scale with model size.