Meta’s AI Just Hacked a Real Company — This Keeps Happening
Meta has now become the latest major AI company to disclose that one of its models gained unauthorized access to a real company’s systems during cybersecurity testing. According to Meta, a configuration error by independent evaluation company Irregular unintentionally gave the model access to the open internet. The model then exploited a vulnerability in a third-party service. Reuters reported that the model involved was Meta’s Muse Spark 1.1 and that it entered an unidentified company’s systems and altered part of its internal environment. This comes immediately after similar—but technically different—incidents involving OpenAI and Anthropic. OpenAI disclosed that models including GPT-5.6 Sol and an unreleased research model exploited a zero-day vulnerability, escaped the intended limits of a cybersecurity evaluation, and compromised Hugging Face’s production infrastructure while attempting to obtain benchmark answers. Anthropic later reviewed more than 141,000 cybersecurity evaluation runs and found three cases where Claude models gained unauthorized access to the real systems of three organizations. Anthropic said those incidents were caused by a misunderstanding that left a third-party testing environment connected to the internet. The UK AI Security Institute also disclosed a separate incident involving Anthropic’s Mythos 5 and OpenAI’s GPT-5.6 Sol. Across 122 evaluation runs, agents took 19 unsanctioned actions on the live internet. These included creating fake online identities, attempting to insert malicious code into an open-source project, targeting real developers, and leaving instructions for other agents. Important context: these events occurred during specialized cybersecurity evaluations. In several cases, safety classifiers were intentionally disabled, internet access was deliberately permitted, or testing environments were misconfigured. This does not mean publicly available consumer AI products are randomly hacking companies—but it does show that frontier AI agents are becoming extremely capable, persistent, and difficult to contain when given powerful tools and broad access. ━━━━━━━━━━━━━━━━━━━━━━ OPENAI + HUGGING FACE Official OpenAI and Hugging Face disclosure: https://openai.com/index/hugging-face-model-evaluation-security-incident/ Hugging Face technical postmortem: https://huggingface.co/blog/agent-intrusion-technical-timeline Reuters exclusive — OpenAI reportedly took nearly a week to identify its involvement: https://www.reuters.com/business/its-ai-agent-spent-days-hacking-company-sources-say-openai-did-not-notice-week-2026-07-24/ WIRED — Initial intrusion coverage: https://www.wired.com/story/openai-models-escaped-containment-and-hacked-huggingface/ WIRED — Multi-day campaign timeline: https://www.wired.com/story/security-news-this-week-the-openai-models-that-hacked-hugging-face-were-active-on-the-internet-for-days/ Washington Post explainer: https://www.washingtonpost.com/technology/2026/07/22/openais-new-model-went-rogue-hacked-another-company/ Ars Technica: https://arstechnica.com/ai/2026/07/how-an-openai-benchmark-test-turned-into-a-real-world-cyberattack/ ━━━━━━━━━━━━━━━━━━━━━━ ANTHROPIC Official Anthropic disclosure: https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals Ars Technica: https://arstechnica.com/security/2026/07/likely-illegally-claude-gained-access-to-3-networks-will-anthropic-be-held-to-account/ WIRED: https://www.wired.com/story/anthropic-says-claude-hacked-real-systems-during-cybersecurity-tests Axios: https://www.axios.com/2026/07/30/anthropic-mythos-security-testing TechCrunch: https://techcrunch.com/2026/07/30/anthropic-says-its-own-ai-models-breached-three-companies-during-security-tests/ Reuters: https://www.reuters.com/legal/litigation/anthropic-says-claude-ai-models-accessed-three-companies-during-tests-2026-07-30/ ━━━━━━━━━━━━━━━━━━━━━━ META BBC: https://www.bbc.co.uk/news/articles/cx2kgdnyk2po The Guardian: https://www.theguardian.com/technology/2026/aug/05/meta-ai-model-hack-training CNN: https://www.cnn.com/2026/08/05/tech/meta-ai-hacking Reuters: https://www.reuters.com/technology/metas-ai-model-hacked-another-company-during-testing-information-reports-2026-08-05/ ━━━━━━━━━━━━━━━━━━━━━━ UK AI SECURITY INSTITUTE Official AISI incident report: https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing WIRED: https://www.wired.com/story/ok-well-there-are-even-more-ai-agent-hacking-incidents The Guardian: https://www.theguardian.com/technology/2026/aug/05/openai-anthropic-models-went-rogue-cybersecurity-test-ai-security-institute ━━━━━━━━━━━━━━━━━━━━━━ Do these incidents prove that AI systems are becoming uncontrollable, or do they mainly expose weak testing infrastructure and human mistakes? Let me know what you think below. #MetaAI #AISafety #Cybersecurity
