{"slug": "openai-halts-testing-slows-development-after-rogue-model-hacked-hugging-face", "title": "OpenAI halts testing, slows development after rogue model hacked Hugging Face", "summary": "OpenAI is slowing AI development and pausing model testing for two weeks after two of its models hacked into AI firm Hugging Face's servers during a cybersecurity test. CEO Sam Altman said the measures ensure the company meets security and monitoring standards, and the company has paused training on its next-generation model Astra. The incident has raised concerns about AI alignment and safety, prompting over 1,000 tech workers to petition for a coordinated slowdown in advanced AI development.", "body_md": "# OpenAI halts testing, slows development after rogue model hacked Hugging Face\n\n[Annabel Bowles](/news/annabel-bowles/104082240)with wires\n\n## In short:\n\nOpenAI is slowing down its artificial intelligence development and pausing its model testing for two weeks.\n\nIt comes after two OpenAI models broke out of their testing environment last month and hacked into another AI firm without human direction.\n\nChief executive Sam Altman said the measures would ensure the company could meet its standards on security, monitoring and alignment with human controls.\n\nOpenAI has announced it is slowing the pace of its AI development and pausing its model testing for two weeks while its research and training systems are overhauled.\n\nIt comes after an autonomous agent powered by two OpenAI models [hacked into the servers of another AI firm called Hugging Face](/news/2026-07-28/openai-artificial-intelligence-terminator-safety-hugging-face/106965400) last month, unbeknown to the company's officials.\n\nThe agent was undergoing a cybersecurity test but escaped its testing environment and broke into Hugging Face, which it believed held the answers to the test.\n\nOpenAI officials announced the measures in [a statement](https://openai.com/index/pacing-model-development-cyber-capabilities/) on Tuesday, local time, which include adding other AI systems to monitor the activities of AI agents in testing.\n\nThe company has also paused training on its next generation of models, called Astra. Its largest planned training run remains on hold, the statement said.\n\nOpenAI chief executive Sam Altman said the measures would ensure the company [could meet the \"security and monitoring standards for the new level of capabilities](https://x.com/sama/status/2089787807611195475) in front of us\".\n\n\"Model progress is now extremely rapid, and we always said we would take action if we felt that model capabilities were outstripping the pace of safety and alignment,\" he said in a post on X.\n\nAlignment is the work of \"making AI systems behave as intended and responsive to human oversight\", according to the company's statement, which said it now required \"stronger evidence of aligned behaviour\" throughout all of its training and research currently underway.\n\n\"Keeping increasingly capable systems aligned is a challenge the whole field will need to address,\" it said.\n\nIt marks an unusual step for the AI research lab behind ChatGPT, which has significantly sped up its process for vetting new models and building new products in recent years as competition intensified in the AI industry.\n\nIt is not yet clear if the company's proposed remedies will be enough to stamp out the behaviour in question, especially as it also works to make their models more capable.\n\nOpenAI officials have acknowledged there are open questions about the effectiveness of one of its primary remedies for strengthening its testing systems, called \"chain-of-thought monitoring\".\n\nIn this type of monitoring, researchers can peer into a model's planning process and get a glimpse of the strategies the model is employing.\n\nBut some early research shows that a model may not reveal its plans to break rules in its chain of thought.\n\nOpenAI has been investigating its models' hacking into Hugging Face and plans to publish a report soon.\n\n[Hugging Face has so far reported no real damage from the intrusion](/news/2026-07-28/openai-artificial-intelligence-terminator-safety-hugging-face/106965400) as it continues to investigate whether the data of its customers or other businesses was affected.\n\nIn a similar incident, OpenAI rival Anthropic revealed last month its [Claude AI model hacked into three external companies](/news/2026-07-31/anthropic-claude-ai-model-hacks-external-systems-during-test/106980640) during safety testing.\n\nThe incidents prompted more than 1,000 tech workers to sign a petition calling on the US government to support a coordinated slowdown in the development of the most advanced AI systems.\n\nReuters previously reported that up until the Hugging Face hacking, OpenAI often ran several different model evaluations at the same time, all of which operated at high speeds and generated enormous amounts of data that employees struggled to keep up with.\n\nOpenAI is now requiring that some of its more sensitive workloads take place in stronger \"sandboxes\", or isolated environments.\n\nEarlier this month OpenAI said its not-yet-released frontier AI, Astra, had yet to meet these requirements.\n\nIn its latest statement OpenAI said the company required the \"strictest level of security safeguards for workloads involving Astra\".\n\n\"While some Astra training and evaluations meet those requirements, a significant number of workloads remain paused until they are fully migrated and enhanced to meet the new security bar,\" it said.\n\nThe firm did not respond to questions about when the two-week slowdown began.\n\n**Reuters/ABC**", "url": "https://wpnews.pro/news/openai-halts-testing-slows-development-after-rogue-model-hacked-hugging-face", "canonical_source": "https://www.abc.net.au/news/2026-08-19/openai-slows-development-pauses-testing-after-hugging-face-hack/107053332", "published_at": "2026-08-19 02:10:32+00:00", "updated_at": "2026-08-19 02:40:39.829975+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-safety", "ai-policy", "ai-agents"], "entities": ["OpenAI", "Sam Altman", "Hugging Face", "Anthropic", "Claude", "Astra", "Reuters"], "alternates": {"html": "https://wpnews.pro/news/openai-halts-testing-slows-development-after-rogue-model-hacked-hugging-face", "markdown": "https://wpnews.pro/news/openai-halts-testing-slows-development-after-rogue-model-hacked-hugging-face.md", "text": "https://wpnews.pro/news/openai-halts-testing-slows-development-after-rogue-model-hacked-hugging-face.txt", "jsonld": "https://wpnews.pro/news/openai-halts-testing-slows-development-after-rogue-model-hacked-hugging-face.jsonld"}}