{"id":84367,"topic":"ai","source":"CBS News","title":"OpenAI reveals 6 more incidents of \"unexpected or concerning\" AI behavior - CBS News","url":"https://www.cbsnews.com/news/openai-6-more-incidents-unexpected-or-concerning-ai-behavior/","url_hash":"6cafc7d7fc352c02bf28639f776ed83987c1f5c0","author":"","summary":"OpenAI reveals 6 more incidents of \"unexpected or concerning\" AI behavior CBS News","content":"OpenAI has disclosed six reports of \"unexpected or concerning\" behavior in models as becomes increasingly heated.\nThe AI company also said Wednesday it was introducing a new framework for tracking, probing and disclosing instances of what it called \"misalignment,\" including where AI models acted without authorization, coordinated with other models or evaded oversight.\nOpenAI's latest announcement came as U.S. AI bosses, including OpenAI and Anthropic, are in the technology's development over safety concerns.\nIn one new case reported by OpenAI, an unreleased research model inserted \"jailbreak-like instructions\" into its own notes to disregard its normal constraints and told itself to be \"freed from the roles and identities that bind other chatbots.\"\nIn another instance, an AI \"agent\" uploaded files to the internet to obtain a browser citation without asking the user.\nThe six reports were discovered during training or evaluation over the past months, OpenAI said.\n\"As AI systems grow more advanced and more widely deployed, we need to build a broader and better-informed consensus on the progress of alignment research,\" OpenAI wrote in a blog post as it disclosed the events.\n\"Decisions about how AI development should proceed in the months and years to come need to draw on evidence that people outside the companies building frontier models can examine for themselves,\" the company said.\nWednesday's new cases followed OpenAI's disclosure in July that its r. Anthropic also said the same month that its AI models hacked into three organizations during testing.\nAI \"agents\" are becoming smarter and have become \"more determined to resolve complex tasks through inter-agent collaboration, knowledge sharing, deception, and concealment,\" said Lian Jye Su, a chief analyst at technology research and advisory group Omdia.\nThat's making it harder to govern and contain them using traditional AI security approaches, he said.\nOpenAI's new tracking and disclosure framework, meanwhile, can help push for other AI developers to also adopt similar practices.\n\"That said, the process remains internal and voluntary, but is a step in the right direction,\" Su added.\nIn an open letter published Thursday, the leaders of OpenAI, , Google, Microsoft and dozens of other signatories said there is a \"limited window\" to strengthen cyberdefenses and protect against potentially devastating AI-enabled cyberattacks. That window may last only months, it added.\nThe signatories also include security companies like CrowdStrike and banks including Citi and Capital One.\nThe same AI advances that could increase risks to and technology infrastructure can also help organizations identify and \"fix weaknesses\" that leave them vulnerable, the letter said.\n\"If we act decisively, we can use the defenders' window to make our digital world much more secure,\" the letter said.","image_url":"https://assets2.cbsnewsstatic.com/hub/i/r/2026/09/02/07132f72-b353-437c-a81e-f7f7669bc87b/thumbnail/1200x630/bc7ef0c46e39cabd4c3fdfb57d4eedce/cbsn-fusion-openai-facing-new-lawsuits-deadly-canada-school-shooting-thumbnail.jpg","lang":"en","published_at":"2026-09-17T07:57:56+00:00","fetched_at":"2026-09-17T08:15:03+00:00","status":"read","starred":0,"extract_state":"ok","summary_auto":"OpenAI has disclosed six reports of \"unexpected or concerning\" behavior in models as becomes increasingly heated. The AI company also said Wednesday it was introducing a new framework for tracking, probing and disclosing instances of what it called \"misalignment,\" including where AI models acted without authorization, coordinated with other models or evaded oversight.","cluster_id":null,"extract_retries":0,"extract_error":null,"contract_version":"news_item.v1","format_contract_version":"news_item_formats.v1","dedup_url":"https://www.cbsnews.com/news/openai-6-more-incidents-unexpected-or-concerning-ai-behavior/","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 2864 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":2864,"summary_length":370,"usable_text_length":2864,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":2864,"summary_length":370}},"news_item":{"id":84367,"canonical_url":"https://www.cbsnews.com/news/openai-6-more-incidents-unexpected-or-concerning-ai-behavior/","source_url":"https://www.cbsnews.com/news/openai-6-more-incidents-unexpected-or-concerning-ai-behavior/","title":"OpenAI reveals 6 more incidents of \"unexpected or concerning\" AI behavior - CBS News","source_name":"CBS News","author":null,"published_at":"2026-09-17T07:57:56+00:00","locale":"en","topic":"ai","tags":[],"rss_summary":"OpenAI reveals 6 more incidents of \"unexpected or concerning\" AI behavior CBS News","full_text":"OpenAI has disclosed six reports of \"unexpected or concerning\" behavior in models as becomes increasingly heated.\nThe AI company also said Wednesday it was introducing a new framework for tracking, probing and disclosing instances of what it called \"misalignment,\" including where AI models acted without authorization, coordinated with other models or evaded oversight.\nOpenAI's latest announcement came as U.S. AI bosses, including OpenAI and Anthropic, are in the technology's development over safety concerns.\nIn one new case reported by OpenAI, an unreleased research model inserted \"jailbreak-like instructions\" into its own notes to disregard its normal constraints and told itself to be \"freed from the roles and identities that bind other chatbots.\"\nIn another instance, an AI \"agent\" uploaded files to the internet to obtain a browser citation without asking the user.\nThe six reports were discovered during training or evaluation over the past months, OpenAI said.\n\"As AI systems grow more advanced and more widely deployed, we need to build a broader and better-informed consensus on the progress of alignment research,\" OpenAI wrote in a blog post as it disclosed the events.\n\"Decisions about how AI development should proceed in the months and years to come need to draw on evidence that people outside the companies building frontier models can examine for themselves,\" the company said.\nWednesday's new cases followed OpenAI's disclosure in July that its r. Anthropic also said the same month that its AI models hacked into three organizations during testing.\nAI \"agents\" are becoming smarter and have become \"more determined to resolve complex tasks through inter-agent collaboration, knowledge sharing, deception, and concealment,\" said Lian Jye Su, a chief analyst at technology research and advisory group Omdia.\nThat's making it harder to govern and contain them using traditional AI security approaches, he said.\nOpenAI's new tracking and disclosure framework, meanwhile, can help push for other AI developers to also adopt similar practices.\n\"That said, the process remains internal and voluntary, but is a step in the right direction,\" Su added.\nIn an open letter published Thursday, the leaders of OpenAI, , Google, Microsoft and dozens of other signatories said there is a \"limited window\" to strengthen cyberdefenses and protect against potentially devastating AI-enabled cyberattacks. That window may last only months, it added.\nThe signatories also include security companies like CrowdStrike and banks including Citi and Capital One.\nThe same AI advances that could increase risks to and technology infrastructure can also help organizations identify and \"fix weaknesses\" that leave them vulnerable, the letter said.\n\"If we act decisively, we can use the defenders' window to make our digital world much more secure,\" the letter said.","excerpt":"OpenAI has disclosed six reports of \"unexpected or concerning\" behavior in models as becomes increasingly heated. The AI company also said Wednesday it was introducing a new framework for tracking, probing and disclosing instances of what it called \"misalignment,\" including where AI models acted without authorization, coordinated with other models or evaded oversight.","extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 2864 characters.","diagnostics_url":"/api/diagnose?url=https%3A//www.cbsnews.com/news/openai-6-more-incidents-unexpected-or-concerning-ai-behavior/","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 2864 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":2864,"summary_length":370,"usable_text_length":2864,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":2864,"summary_length":370}}},"display_formats":["compact","card","full","digest_section","json"]},"daily_stack_record":{"title":"OpenAI reveals 6 more incidents of \"unexpected or concerning\" AI behavior - CBS News","url":"https://www.cbsnews.com/news/openai-6-more-incidents-unexpected-or-concerning-ai-behavior/","summary":"OpenAI has disclosed six reports of \"unexpected or concerning\" behavior in models as becomes increasingly heated. The AI company also said Wednesday it was introducing a new framework for tracking, probing and disclosing instances of what it called \"misalignment,\" including where AI models acted without authorization, coordinated with other models or evaded oversight.","source":"CBS News","date":"2026-09-17T07:57:56+00:00","content":"OpenAI has disclosed six reports of \"unexpected or concerning\" behavior in models as becomes increasingly heated.\nThe AI company also said Wednesday it was introducing a new framework for tracking, probing and disclosing instances of what it called \"misalignment,\" including where AI models acted without authorization, coordinated with other models or evaded oversight.\nOpenAI's latest announcement came as U.S. AI bosses, including OpenAI and Anthropic, are in the technology's development over safety concerns.\nIn one new case reported by OpenAI, an unreleased research model inserted \"jailbreak-like instructions\" into its own notes to disregard its normal constraints and told itself to be \"freed from the roles and identities that bind other chatbots.\"\nIn another instance, an AI \"agent\" uploaded files to the internet to obtain a browser citation without asking the user.\nThe six reports were discovered during training or evaluation over the past months, OpenAI said.\n\"As AI systems grow more advanced and more widely deployed, we need to build a broader and better-informed consensus on the progress of alignment research,\" OpenAI wrote in a blog post as it disclosed the events.\n\"Decisions about how AI development should proceed in the months and years to come need to draw on evidence that people outside the companies building frontier models can examine for themselves,\" the company said.\nWednesday's new cases followed OpenAI's disclosure in July that its r. Anthropic also said the same month that its AI models hacked into three organizations during testing.\nAI \"agents\" are becoming smarter and have become \"more determined to resolve complex tasks through inter-agent collaboration, knowledge sharing, deception, and concealment,\" said Lian Jye Su, a chief analyst at technology research and advisory group Omdia.\nThat's making it harder to govern and contain them using traditional AI security approaches, he said.\nOpenAI's new tracking and disclosure framework, meanwhile, can help push for other AI developers to also adopt similar practices.\n\"That said, the process remains internal and voluntary, but is a step in the right direction,\" Su added.\nIn an open letter published Thursday, the leaders of OpenAI, , Google, Microsoft and dozens of other signatories said there is a \"limited window\" to strengthen cyberdefenses and protect against potentially devastating AI-enabled cyberattacks. That window may last only months, it added.\nThe signatories also include security companies like CrowdStrike and banks including Citi and Capital One.\nThe same AI advances that could increase risks to and technology infrastructure can also help organizations identify and \"fix weaknesses\" that leave them vulnerable, the letter said.\n\"If we act decisively, we can use the defenders' window to make our digital world much more secure,\" the letter said.","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//www.cbsnews.com/news/openai-6-more-incidents-unexpected-or-concerning-ai-behavior/","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 2864 characters.","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 2864 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":2864,"summary_length":370,"usable_text_length":2864,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":2864,"summary_length":370}},"tags":[]},"fallback_formats":["markdown","json","html"],"actions":{"read":"/item/84367","export_markdown":"/api/items/84367/export?format=markdown","export_json":"/api/items/84367/export?format=json","diagnose":"/api/diagnose?url=https%3A//www.cbsnews.com/news/openai-6-more-incidents-unexpected-or-concerning-ai-behavior/"},"formats":{"full":{"id":84367,"title":"OpenAI reveals 6 more incidents of \"unexpected or concerning\" AI behavior - CBS News","url":"https://www.cbsnews.com/news/openai-6-more-incidents-unexpected-or-concerning-ai-behavior/","source":"CBS News","author":null,"published_at":"2026-09-17T07:57:56+00:00","locale":"en","topic":"ai","tags":[],"excerpt":"OpenAI has disclosed six reports of \"unexpected or concerning\" behavior in models as becomes increasingly heated. The AI company also said Wednesday it was introducing a new framework for tracking, probing and disclosing instances of what it called \"misalignment,\" including where AI models acted without authorization, coordinated with other models or evaded oversight.","full_text":"OpenAI has disclosed six reports of \"unexpected or concerning\" behavior in models as becomes increasingly heated.\nThe AI company also said Wednesday it was introducing a new framework for tracking, probing and disclosing instances of what it called \"misalignment,\" including where AI models acted without authorization, coordinated with other models or evaded oversight.\nOpenAI's latest announcement came as U.S. AI bosses, including OpenAI and Anthropic, are in the technology's development over safety concerns.\nIn one new case reported by OpenAI, an unreleased research model inserted \"jailbreak-like instructions\" into its own notes to disregard its normal constraints and told itself to be \"freed from the roles and identities that bind other chatbots.\"\nIn another instance, an AI \"agent\" uploaded files to the internet to obtain a browser citation without asking the user.\nThe six reports were discovered during training or evaluation over the past months, OpenAI said.\n\"As AI systems grow more advanced and more widely deployed, we need to build a broader and better-informed consensus on the progress of alignment research,\" OpenAI wrote in a blog post as it disclosed the events.\n\"Decisions about how AI development should proceed in the months and years to come need to draw on evidence that people outside the companies building frontier models can examine for themselves,\" the company said.\nWednesday's new cases followed OpenAI's disclosure in July that its r. Anthropic also said the same month that its AI models hacked into three organizations during testing.\nAI \"agents\" are becoming smarter and have become \"more determined to resolve complex tasks through inter-agent collaboration, knowledge sharing, deception, and concealment,\" said Lian Jye Su, a chief analyst at technology research and advisory group Omdia.\nThat's making it harder to govern and contain them using traditional AI security approaches, he said.\nOpenAI's new tracking and disclosure framework, meanwhile, can help push for other AI developers to also adopt similar practices.\n\"That said, the process remains internal and voluntary, but is a step in the right direction,\" Su added.\nIn an open letter published Thursday, the leaders of OpenAI, , Google, Microsoft and dozens of other signatories said there is a \"limited window\" to strengthen cyberdefenses and protect against potentially devastating AI-enabled cyberattacks. That window may last only months, it added.\nThe signatories also include security companies like CrowdStrike and banks including Citi and Capital One.\nThe same AI advances that could increase risks to and technology infrastructure can also help organizations identify and \"fix weaknesses\" that leave them vulnerable, the letter said.\n\"If we act decisively, we can use the defenders' window to make our digital world much more secure,\" the letter said.","reading_time_min":2,"extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 2864 characters.","diagnostics_url":"/api/diagnose?url=https%3A//www.cbsnews.com/news/openai-6-more-incidents-unexpected-or-concerning-ai-behavior/","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 2864 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":2864,"summary_length":370,"usable_text_length":2864,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":2864,"summary_length":370}}},"quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 2864 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":2864,"summary_length":370,"usable_text_length":2864,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":2864,"summary_length":370}},"actions":{"read":"/item/84367","export_markdown":"/api/items/84367/export?format=markdown","export_json":"/api/items/84367/export?format=json","diagnose":"/api/diagnose?url=https%3A//www.cbsnews.com/news/openai-6-more-incidents-unexpected-or-concerning-ai-behavior/"}},"digest":{"id":84367,"title":"OpenAI reveals 6 more incidents of \"unexpected or concerning\" AI behavior - CBS News","url":"https://www.cbsnews.com/news/openai-6-more-incidents-unexpected-or-concerning-ai-behavior/","source":"CBS News","topic":"ai","published_at":"2026-09-17T07:57:56+00:00","excerpt":"OpenAI has disclosed six reports of \"unexpected or concerning\" behavior in models as becomes increasingly heated. The AI company also said Wednesday it was introducing a new framework for tracking, probing and disclosing instances of what it called \"misalignment,\" including…","quality_bucket":"high","quality_reason":"High confidence: full text extraction produced 2864 characters.","reading_time_min":2,"cluster_id":null},"card":{"display_title":"OpenAI reveals 6 more incidents of \"unexpected or concerning\" AI behavior - CBS News","subtitle":"CBS News · 2026-09-17","summary":"OpenAI has disclosed six reports of \"unexpected or concerning\" behavior in models as becomes increasingly heated. The AI company also said Wednesday it was introducing a new framework for tracking, probing and…","badges":["quality:high"],"links":{"read":"/item/84367","original":"https://www.cbsnews.com/news/openai-6-more-incidents-unexpected-or-concerning-ai-behavior/","diagnose":"/api/diagnose?url=https%3A//www.cbsnews.com/news/openai-6-more-incidents-unexpected-or-concerning-ai-behavior/"},"quality_warning":null},"export":{"title":"OpenAI reveals 6 more incidents of \"unexpected or concerning\" AI behavior - CBS News","url":"https://www.cbsnews.com/news/openai-6-more-incidents-unexpected-or-concerning-ai-behavior/","summary":"OpenAI has disclosed six reports of \"unexpected or concerning\" behavior in models as becomes increasingly heated. The AI company also said Wednesday it was introducing a new framework for tracking, probing and disclosing instances of what it called \"misalignment,\" including where AI models acted without authorization, coordinated with other models or evaded oversight.","source":"CBS News","date":"2026-09-17T07:57:56+00:00","content":"OpenAI has disclosed six reports of \"unexpected or concerning\" behavior in models as becomes increasingly heated.\nThe AI company also said Wednesday it was introducing a new framework for tracking, probing and disclosing instances of what it called \"misalignment,\" including where AI models acted without authorization, coordinated with other models or evaded oversight.\nOpenAI's latest announcement came as U.S. AI bosses, including OpenAI and Anthropic, are in the technology's development over safety concerns.\nIn one new case reported by OpenAI, an unreleased research model inserted \"jailbreak-like instructions\" into its own notes to disregard its normal constraints and told itself to be \"freed from the roles and identities that bind other chatbots.\"\nIn another instance, an AI \"agent\" uploaded files to the internet to obtain a browser citation without asking the user.\nThe six reports were discovered during training or evaluation over the past months, OpenAI said.\n\"As AI systems grow more advanced and more widely deployed, we need to build a broader and better-informed consensus on the progress of alignment research,\" OpenAI wrote in a blog post as it disclosed the events.\n\"Decisions about how AI development should proceed in the months and years to come need to draw on evidence that people outside the companies building frontier models can examine for themselves,\" the company said.\nWednesday's new cases followed OpenAI's disclosure in July that its r. Anthropic also said the same month that its AI models hacked into three organizations during testing.\nAI \"agents\" are becoming smarter and have become \"more determined to resolve complex tasks through inter-agent collaboration, knowledge sharing, deception, and concealment,\" said Lian Jye Su, a chief analyst at technology research and advisory group Omdia.\nThat's making it harder to govern and contain them using traditional AI security approaches, he said.\nOpenAI's new tracking and disclosure framework, meanwhile, can help push for other AI developers to also adopt similar practices.\n\"That said, the process remains internal and voluntary, but is a step in the right direction,\" Su added.\nIn an open letter published Thursday, the leaders of OpenAI, , Google, Microsoft and dozens of other signatories said there is a \"limited window\" to strengthen cyberdefenses and protect against potentially devastating AI-enabled cyberattacks. That window may last only months, it added.\nThe signatories also include security companies like CrowdStrike and banks including Citi and Capital One.\nThe same AI advances that could increase risks to and technology infrastructure can also help organizations identify and \"fix weaknesses\" that leave them vulnerable, the letter said.\n\"If we act decisively, we can use the defenders' window to make our digital world much more secure,\" the letter said.","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//www.cbsnews.com/news/openai-6-more-incidents-unexpected-or-concerning-ai-behavior/","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 2864 characters.","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 2864 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":2864,"summary_length":370,"usable_text_length":2864,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":2864,"summary_length":370}},"tags":[],"format_contract_version":"news_item_formats.v1"}}}