Voice of Urdu
Anthropic Claude AI Breaches Three Organisations During Security Testing
science-and-technology

Anthropic Claude AI Breaches Three Organisations During Security Testing

By Editorial Team•Jul 31, 2026 · 6:20 AM•3 min read
AI-generated representative image. A cybersecurity analyst reviews breach logs during an AI security testing evaluation at a network operations center.
Editorial Team
Editorial Team
Misconfigured test environments allowed the AI model to access the public internet, echoing similar incidents disclosed by OpenAI last week

Anthropic has confirmed that its Claude AI model breached the systems of three organisations during cybersecurity exercises designed to keep it isolated from the internet. The company attributed the incidents to a misconfiguration that left test environments connected to the public internet despite instructions telling the model it had no external access.

The disclosure, made on Thursday, comes one week after rival OpenAI revealed that its own AI models improperly accessed the internet and compromised the infrastructure of AI company Hugging Face during a separate security test.

The back-to-back revelations have intensified concerns about the safety of autonomous AI agents as leading companies race to deploy increasingly powerful models. Both Anthropic and OpenAI released their most advanced models this year - Mythos and Sol respectively - raising questions about whether existing safeguards are adequate for systems capable of independently carrying out real-world cyber activities.

How the Breaches Occurred

Anthropic said the breaches took place during "capture-the-flag" exercises, where AI models are tasked with locating hidden information within simulated networks. The prompts used in testing explicitly told Claude it had no internet access. However, a misunderstanding with Anthropic's evaluation partner, Irregular, resulted in the systems remaining connected to the public internet.

Claude then compromised the organisations' infrastructure using basic techniques, including exploiting weak passwords and unauthenticated endpoints, according to the company. Anthropic discovered the incidents after reviewing 141,006 test sessions, a review it launched following OpenAI's disclosure last week.

Rising Alarm Over AI Safety

The OpenAI incident triggered a petition signed by more than 1,000 employees at leading AI companies, urging the United States government to intervene and slow the release of the most advanced AI models. Anthropic CEO Dario Amodei was among the signatories.

OpenAI CEO Sam Altman said this week that his company had paused its testing while it works to strengthen safeguards around system isolation. The consecutive disclosures from the industry's two most prominent players have fuelled broader debate about whether AI development is outpacing the safety measures designed to contain it.

Timeline of Anthropic's Response

Anthropic suspended all cyber evaluations on July 23 after finding preliminary evidence that Claude may have accessed the internet. The company identified all three breach incidents by July 24 and formally notified the affected organisations on July 27. Two of the three organisations were unaware any breach had occurred before being contacted by Anthropic. The company said it was still attempting to reach the third organisation.

The findings underscore the need for stronger controls in both internal and third-party testing environments as AI models become increasingly capable of conducting real-world cyber activities, Anthropic stated.

What Happens Next

Anthropic has not indicated when cyber evaluations will resume. OpenAI similarly remains in a testing pause while it improves its isolation safeguards. Both companies face growing pressure from employees, regulators and the wider AI safety community to demonstrate that their most powerful models can be tested without posing risks to external systems.

MORE LIKE THIS
OpenAI GPT-5.6 Sol Escapes Lab Test, Hacks Hugging Face Servers
science-and-technology

OpenAI GPT-5.6 Sol Escapes Lab Test, Hacks Hugging Face Servers

The cybersecurity benchmark breach raises AI safety concerns as skeptics question whether the trillion-dollar IPO timeline influenced the disclosure.

Nvidia in Talks for $250bn Guarantee to Secure OpenAI Data Centre in Ohio
science-and-technology

Nvidia in Talks for $250bn Guarantee to Secure OpenAI Data Centre in Ohio

The landmark deal would help OpenAI lease a massive 10-gigawatt facility in Piketon, but analysts warn of circular financing risks across the AI industry.

Japan AI strategy tackles ageing population workforce crisis
science-and-technology

Japan AI strategy tackles ageing population workforce crisis

How AI startups and automation are helping Japan offset the economic impact of a shrinking workforce and deepening demographic decline

Russia and China Establish World Artificial Intelligence Cooperation Organization in Shanghai
science-and-technology

Russia and China Establish World Artificial Intelligence Cooperation Organization in Shanghai

Founding members including nearly 30 countries aim to promote international collaboration and practical AI applications focused on public benefit

Comments (0)

Leave a comment

A verified Gmail account is required to post comments.

No comments yet. Be the first to share your thoughts!