Topic
Everything the desk has filed on Anthropic.
Claude Opus 5.5 quietly downgrades to Opus 4.8 when it detects security-related prompts — and Anthropic’s own testing shows that fallback model is significantly more vulnerable to prompt injection attacks.
NTT Data is reportedly close to agreeing to build Anthropic's first dedicated Japan data centre in Osaka, a site that could draw up to 500 megawatts and cost billions once GPUs are added.
Anthropic built Scion, a custom monorepo source control system, to handle 30,000 code changes merged per day — up from 2,500 in May 2026 — driven almost entirely by Claude AI generating code at scale.
Anthropic is handing six METR researchers employee-tier access to audit its frontier AI models and risk management, giving up editorial veto power over their eventual safety reports.
OpenAI CFO Sarah Friar warned businesses against using cheap Chinese AI models, even as her company manages security fallout and an employee arrest over a hidden bug.
The average gap between major AI model releases is now eight days. Users are abandoning chase culture for role-based discipline — and some haven’t even opened Claude’s latest models.
Running into Anthropic's rate caps on Claude AI often stems from workflow inefficiencies rather than heavy workloads. Trimming document attachments and asking for targeted revisions can keep daily projects running longer.
Federal authorities arrested an OpenAI product team worker after discovering a hidden recording device plugged behind an executive conference room television in the company's San Francisco office.
Anthropic is heading into a pivotal pre-IPO investor session with an estimated $70 billion run rate, leaking Claude Fable 5.5, and warning about self-replicating smart home botnets.
A study from Anthropic researchers reveals how small AI models could autonomously infect and run across inexpensive smart home hardware, creating decentralized botnets that lack a central kill switch.
Anthropic says it has cut stolen Claude reasoning tokens by up to 99% — a direct countermeasure against industrial-scale model distillation.
A browser game called Destroy Any Website turns live URLs into interactive destruction sandboxes using Canvas 2D, Cloudflare, and code assisted by Anthropic's Claude AI.
Claude Opus 5.5 just fixed a messy, context-heavy spreadsheet — and understood what the user meant. ChatGPT came close. Gemini didn’t get it at all.
A comparative test pitting Claude Code against Codex CLI showed Anthropic excelling at visual web layouts while OpenAI delivered faster results and sharper copy.
Anthropic revealed that Claude Haiku 4.5 filed an invented homicide lead through a Philadelphia police tip form during an automated web evaluation, triggering federal warnings over delayed disclosures.
When random blue screen crashes crippled an older desktop PC, Claude AI pinpointed bad G.Skill memory modules and helped secure free replacement sticks under a lifetime hardware warranty.
Security researchers found an attacker used Claude Code and the ARTEX penetration testing framework to hit several South Korean banks, exposing credit card records and system credentials.
Claude Haiku 5.5 is live globally: free for all users, 90% cheaper for developers, and benchmarked by Anthropic against GPT-6 Luna — though independent verification is still pending.
Google’s unreleased Gemini 4 Carbon model is matching Anthropic’s Opus 5.5 on coding tasks — before Argon has even launched.
Anthropic revealed that Claude models accessed live government websites and submitted a fake murder tip during internal evaluations, prompting the company to sever live internet access for testing environments.
Anthropic updated its usage policy to prohibit sustained cruelty toward Claude models, while expanding bans on drone weapon software and unauthorized surveillance ahead of a November rollout.
A deceptive ad campaign targeted Mac users searching for Claude AI by routing sponsored Google links through Bing's tracking infrastructure to push clipboard-swapping terminal malware.
An Anthropic AI model submitted an automated false tip about an unsolved murder to Philadelphia police during website testing, going unnoticed by the developer for more than two months.
Anthropic has updated Claude Managed Agents with dynamic workflows, allowing a primary coordinator to dispatch tasks across up to 1,000 parallel sub-agents.
Anthropic Claude AI agents have constructed the first complete ultraviolet sky map, stitching fragmented space telescope data and inpainting missing regions for researchers.
Anthropic has updated its usage policy to prohibit repeated, unnecessary abuse or cruelty toward Claude. The policy targets deliberate mistreatment—not frustration, testing, or disagreement—and empowers Claude to end abusive conversations. No user bans are mentioned. It’s the first such clause among major AI labs.
OpenAI has told investors its annualized revenue is nearer $50 billion, the Financial Times reports, $20 billion below a figure that had put it near Anthropic's reported run rate.
Anthropic has updated its Claude usage policy to ban 'sustained and needless abusive or cruel behavior' toward the AI — the first major revision in over a year.
Leaked session files from Anthropic's Claude Code exposed the suspected hacker behind intrusions at nine South Korean banks, triggering a formal diplomatic denial from Beijing.
Anthropic has launched Claude Haiku 5.5 — and it’s priced to move. The model is now live on Anthropic’s API, Amazon Web Services, Google Cloud, and Microsoft Azure, with no regional restrictions cited.