Anthropic releases Fable and Mythos 5.2 with shared training-time limits
Anthropic says its new Fable and Mythos models share targeted limits on AI-development skills while offering different safeguards and access.
Amazon describes replacing a general AI model with a ticket decision model
Amazon says it switched an internal ticket classifier to a specialized decision model, with substantially lower modeled serving costs.
Notion says Fireworks incident exposed a dataset of AI chat transcripts
Notion says a Fireworks security incident affected an Insights dataset containing AI chat transcripts from fewer than 1% of workspaces.
Google’s Gemini policy permits limited human review of connected-app material
Google says a subset of consumer Gemini interactions derived from connected apps may be human-reviewed, sometimes including a whole relevant file or email.
Google announces confidential-inference and dense-memory TPU systems
Google says TPU 8tx targets confidential workloads while TPU 8i Max combines two chips for greater compute and memory bandwidth.
LingBot-Touch researchers report a shared representation for robot touch
Researchers report a tactile representation trained across sensor types and tested with downstream robot manipulation policies.