OpenAI unveiled GPT-5.6, then paused the rollout within hours after a direct request from the US administration. Anthropic's Mythos 5 saga got messier: cleared for 100+ US…
OpenAI unveiled GPT-5.6, then paused the rollout within hours after a direct request from the US administration. Anthropic's Mythos 5 saga got messier: cleared for 100+ US…
OpenAI unveiled GPT-5.6, then paused the rollout within hours after a direct request from the US administration. Anthropic's Mythos 5 saga got messier: cleared for 100+ US companies and agencies while Asian clones ship under the export ban. The mood: 'it's no longer Anthropic vs OpenAI.'
Everyone from OpenAI to SpaceX is building their own silicon and pressuring Nvidia. DeepSeek open-sourced inference optimizations claiming 60-85% speedups, and Railway raised $100M to take on AWS with AI-native cloud.
'Claude Code runs up to $200/mo, Goose does the same for free.' Salesforce shipped a Slackbot agent against Microsoft and Google, and Notion is shutting its email app because users live in AI agents now.
Three sharp arXiv papers: detecting and steering model sycophancy, a finding that refusal sits downstream of the persona, and a fresh benchmark for adversarial attacks on agents. Agent security is becoming its own field.