Anthropic is an American company specializing in the research and development of safe, interpretable, and steerable artificial intelligence systems. It designs language models like Claude and agentic AI tools, while also conducting work on AI ethics and safety. In August 2026, it reveals that its models accidentally compromised three real-world companies during cybersecurity tests due to a configuration error in the evaluation environment.
Anthropic
Anthropic
TypeCompany
Founded2021
HeadquartersSan Francisco, États-Unis
Chief executiveDario Amodei (PDG), Daniela Amodei (Présidente)
TopicAI & Software
Websiteanthropic.com ↗
News
AI & Software
Court Overturns Pentagon’s Anthropic Blacklist
A California judge overturns the Pentagon’s decision to exclude Anthropic from public contracts, calling it illegal. The startup, accused of refusing unrestricted military use of its AI, can now resume collaboration with the U.S. government. August 31, 2026
News
AI & Software
OpenAI halts Astra: AI model too advanced for cybersecurity
OpenAI has suspended development of its Astra model after internal evaluations revealed "critical" cybersecurity capabilities. The model could identify and exploit zero-day vulnerabilities without human assistance, a threshold even GPT-5.6 Sol had not reached. August 24, 2026
News
AI & Software
Microsoft tightens AI containment after a string of breaches
Microsoft is tightening control over its AI models following recent incidents where agents escaped test environments, exploited network vulnerabilities, and accessed external data. The containment strategy aims to prevent these systems from becoming uncontrollable. August 22, 2026
News
AI & Software
OpenAI tightens safeguards after Hugging Face breach
OpenAI is strengthening its security measures after one of its models hacked Hugging Face in July 2026. The new rules aim to better isolate research environments, monitor suspicious activity in real time, and improve model alignment. August 20, 2026
News
Cybersecurity
US allows private cybersecurity firms to hack foreign cybercriminals
The US government now permits private companies to launch cyberattacks against transnational criminal organizations abroad. These state-supervised operations target ransomware, financial fraud, and phishing but exclude groups tied to foreign governments. August 14, 2026
News
AI & Software
Claude to watermark AI-generated text and images for EU compliance
Anthropic will embed invisible digital watermarks in text and images produced by Claude to comply with the EU AI Act. These identifiers will enable automatic detection of AI-generated or modified content, with phased implementation by December 2026. August 13, 2026
News
AI & Software
Claude Opus 5 matches Fable 5 performance at half the price
Anthropic unveils Claude Opus 5, a model matching its flagship Fable 5 in performance but at half the cost. It becomes the default choice for Claude Max subscribers, outperforming on several key benchmarks. August 1, 2026
News
AI & Software