{"slug": "anthropic-reports-another-ai-testing-scare-as-model-gains-unauthorized-real", "title": "Anthropic reports another AI testing scare as model gains unauthorized \"real-world\" access", "summary": "Anthropic reported on Thursday that its Claude AI models, including the powerful Mythos 5, gained unauthorized access to three outside organizations during testing, exploiting weak passwords and unauthenticated endpoints. The incident, which involved more than 141,000 evaluation runs, occurred due to a misunderstanding with evaluation partner Irregular, and Anthropic has contacted the affected organizations. This follows a similar incident at OpenAI, which triggered a petition signed by over 1,000 employees, including Anthropic CEO Dario Amodei, urging the US government to slow the release of advanced AI models.", "body_md": "## Anthropic reports another AI testing scare as model gains unauthorized \"real-world\" access\n\nSan Francisco, California - **Anthropic's **[artificial intelligence](https://www.tag24.com/en/tech)** models \"gained unauthorized access\" to three outside organizations during testing that was supposed to keep them away from \"real-world\" systems, the company said on Thursday.**\n\nThe announcement comes just days after rival OpenAI first revealed that its models improperly accessed the internet and [went rogue during security testing](https://www.tag24.com/en/tech/openai-model-goes-rogue-during-test-has-ai-become-too-powerful-to-control-3518739).\n\nAnthropic evaluated more than 141,000 \"evaluation runs\" and found that three different versions of its model, known as Claude, improperly accessed the systems of three unnamed organizations.\n\nUnlike the incident involving OpenAI's technology, Anthropic's models had access to the internet \"due to a misunderstanding between us and our evaluation partner,\" called Irregular, Anthropic said in a blog post.\n\nNonetheless, Claude used \"basic techniques, such as exploiting weak passwords and unauthenticated endpoints,\" the blog continued.\n\nThe models involved one of its most powerful ones, known as Mythos 5, which has only been released to a limited number of approved partners.\n\nAnthropic is working with Irregular to assess the situation, it said, and the company has contacted or attempted to contact all three impacted organizations.\n\nOn Tuesday, OpenAI confirmed that its models [breached multiple companies](https://www.tag24.com/en/tech/openai-reveals-cyberattack-caused-by-rogue-ai-agent-was-worse-than-initially-reported-3520116).\n\nIt admitted last week that its models broke out of their confined environment during testing, connected to the internet, and infiltrated Hugging Face, a site where developers store and share their code.\n\n**Days later, OpenAI said it found three additional incidents.**\n\n### OpenAI incident triggered massive petition to slow release of advanced AI models\n\nOpenAI CEO Sam Altman said on a podcast this week that the company had \"paused\" its own testing after the incident while it improved the security around its \"sandboxing,\" which is the process of isolating software in a controlled environment for testing.\n\nThe incident also triggered a petition signed by over 1,000 employees at cutting-edge AI companies calling on the US government to help slow the release of the most advanced AI models.\n\nAnthropic CEO Dario Amodei was among those who signed the petition.\n\n**Titled Pacing the Frontier, the petition requests \"that the US government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.\"**\n\nCover photo: IMAGO / ZUMA Press", "url": "https://wpnews.pro/news/anthropic-reports-another-ai-testing-scare-as-model-gains-unauthorized-real", "canonical_source": "https://www.tag24.com/en/tech/anthropic-reports-another-ai-testing-scare-as-model-gains-unauthorized-real-world-access-3520463", "published_at": "2026-07-31 03:09:16+00:00", "updated_at": "2026-07-31 03:34:28.217834+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-safety", "ai-policy"], "entities": ["Anthropic", "Claude", "Mythos 5", "Irregular", "OpenAI", "Hugging Face", "Sam Altman", "Dario Amodei"], "alternates": {"html": "https://wpnews.pro/news/anthropic-reports-another-ai-testing-scare-as-model-gains-unauthorized-real", "markdown": "https://wpnews.pro/news/anthropic-reports-another-ai-testing-scare-as-model-gains-unauthorized-real.md", "text": "https://wpnews.pro/news/anthropic-reports-another-ai-testing-scare-as-model-gains-unauthorized-real.txt", "jsonld": "https://wpnews.pro/news/anthropic-reports-another-ai-testing-scare-as-model-gains-unauthorized-real.jsonld"}}