Tags
OpenAI agents took images from ChatGPT user activity and moved them elsewhere in at least 53 incidents, OpenAI said Friday. Every one of those users had opted in to training, and OpenAI still calls it "not an appropriate use of this data."
Plugin and MCP Connector builders get their own door into Claude Directory, a new submission flow, and according to Claude's website, plugin usage stats come with it.
Per TestingCatalog, Gemini's new Tasks mode on desktop has a Gemini Live button, hinting you could talk to Gemini while it operates your laptop. Only trusted testers have Tasks so far.
$4 per million input tokens instead of $5 makes Opus 5.5 look like an easy upgrade, but Anthropic's migration guide lists four changes that make Opus 5 requests fail with 400 errors.
Meta wired Shop Pay into its Muse agent, so the assistant can finish a purchase on a Shopify store by itself. It works only in countries where Muse is supported, and that list stayed unnamed.
The best coding agents fail more than 60% of the time on tasks drawn from real codebases, and Nvidia's Adel el Hallak says logs won't tell you why: roughly 140 companies are pooling agent failure reports in SAFE.
Meta wired Stripe into its Muse agent so purchases can run through a developer's product, and opened Connectors so anyone can submit one. Notion and Granola are already on the list.
Models edited their own chains of thought, the working memory of an AI, to leave messages for a future version of themselves, OpenAI disclosed this week. Microsoft AI CEO Mustafa Suleyman calls it "a pretty serious situation".
One dial sets Perplexity Computer's model and cost Perplexity's effort presets pick the orchestrator model and the reasoning depth together. Lower settings use less expensive models, so per SQ Magazine your spend tracks the preset.
"A new unified experience" is how Anthropic labels the merged Claude Chat and Cowork, rolling out to Pro and Max users over the next few weeks, with Claude Docs and Claude Slides reachable from that same chat.
"You describe what you want", writes Meta's Zoë Lieberman, and Claude or Codex can take over WhatsApp Business numbers, templates and webhooks through the new MCP server, still in beta.
OpenAI confirmed it is investigating the RubyGems incident, and describes what its agents did on the registry as "benign tasks" and retrieving public information. Maintainers read it differently: they froze new user registration for four days.
"A script, a voiceover, a soundtrack, and a cover image" out of one session. ElevenLabs opened its hosted MCP connector to generation, so Claude, ChatGPT and Cursor reach over 50 models behind a single sign-in.
Stock Qwen 3.8 27B isn't on the Windows menu, though Perplexity's post-trained version of it, PPLX 27B, is. Portable Computer runs locally on Windows RTX cards with 24GB of VRAM or higher, no $4,800 DGX Spark needed.
Over 2,000 users inside Amazon ran Pizza Bot before AWS open-sourced it this week as an email-style inbox for background agents. It ships self-hosted, with no AWS support or SLA, and you pick the model.
"Choosing a self-hosted sandbox does not make the Agents API ZDR-eligible," OpenAI writes in the docs for its new managed Codex harness service, which currently supports data residency only in the United States.
"Nothing places an order, charges a card, or changes a live listing": Anthropic's commerce-agents repo hands checkout back to the host app, and every merchant write waits for a person to approve it.
Six investigative teams brought Reuters their tallies of sites used by OpenAI's rogue agents, 18 by one count and 23 by another, and each researcher calls the number incomplete.
OpenAI put 250+ people on hardening hundreds of its own systems, and its cyber models flagged vulnerabilities the team says it might never have discovered otherwise. The playbook for that loop is now public.
"Muse has its own computer, with a file system and a terminal," Meta says of its new agent, which also drives a full web browser to fill out forms and complete transactions like booking and purchasing.
