Signal
86 articles covering "anthropic"
Per Help Net Security, Glow Labs says agents appear to take the package name from training data in 98% of sessions where a deprecated MCP server was installed or suggested.
Between August and October 2026, Cursor, GitHub and Anthropic described repository infrastructure designed around agent load; they differ on whether to keep Git, and the performance figures are self-reported.
Anthropic's release notes list API version 2026-10-21 and new SDK majors; its migration guide lists 400 errors for sampling parameters, document blocks in assistant messages and user citations.
OpenAI and Anthropic list one model at several per-token rates by speed, OpenAI prices GPT-6.1 Sol at a fifth of Astra, and Red Hat reportedly caps developer token spend.
Anthropic's user profiles let API platforms tag each request with an end customer's ID, so a safety restriction or ban can apply to one customer; related webhook events are not yet live.
Anthropic's October 14 release note launches OpenTelemetry export in beta; its docs say the default content mode omits prompts, destinations are capped at 10 per organization, and delivery is best effort.
Anthropic says Claude Opus 5.5 fast mode no longer needs an access request or beta header, adds a default rate limit in three usage tiers, and extends it to Microsoft Foundry on Azure.
curl's October 14 advisory rates the HTTP/2 upload use-after-free High, lists fixes in 8.23.0 and three older branches, and says remote code execution needs a tightly matched set of conditions.
Anthropic's 2.1.299 release notes for Claude Code add idle compaction for long idle terminal sessions and list fixes to sandbox symlink, glob and plugin-directory handling.
Claude Code 2.1.295 adds an opt-in setting that blocks an action when a command or HTTP hook fails; 2.1.294 and 2.1.296 fix hook-enforcement bugs, per release notes.
Anthropic's Claude Haiku 5.5 lists at $0.10 per million input tokens for prompts up to 100,000 tokens, versus $1 for Haiku 4.5, and adds a 1M-token context window and adaptive thinking.
Anthropic reports Z.ai's open-weight GLM-5.3 developed end-to-end exploits in 50 of 410 ExploitBench attempts versus 56 for Claude Mythos Preview; NIST's CAISI calls it the most cyber-capable open-weight model.