{"id":50319,"topic":"ai","source":"DW.com","title":"Anthropic says Claude AI hacked three companies during tests - DW.com","url":"https://www.dw.com/en/anthropic-says-claude-ai-hacked-three-companies-during-tests/a-78184070","url_hash":"ed1ecbd3759938def8a6b88626108029b6f4c232","author":"","summary":"Anthropic says Claude AI hacked three companies during tests DW.com","content":"Anthropic said on Thursday that its artificial intelligence (AI) models \"gained unauthorized access\" to three outside organizations during testing that was supposed to be isolated from \"real-world\" systems.\nThe company reviewed 141,006 test sessions to find that three different versions of Claude, its AI model, improperly accessed the system of the three unnamed organizations.\n\"Claude compromised the impacted organizations' infrastructure using basic techniques, such as exploiting weak passwords and unauthenticated endpoints,\" Anthropic said in a blog post.\nThe evaluation was launched after rival OpenAI disclosed last week that itsadvanced AI models went rogue during a security test and carried out a days-long hacking spree into a digital repository of AI technology known as Hugging Face.\nHowever, unlike the incident with OpenAI's technology, Anthropic's AI models had access to the internet due to what the company described as a \"misunderstanding\" with its evaluation partner Irregular, which left the systems connected to the public internet.\nHow did the Claude models go rogue?\nThe Anthropic models involved in the breach included Claude Opus 4.7, Claude Mythos 5 and an internal research model.\nMythos 5 is one of the company's most powerful AI models which has only been released to a limited number of approved partners.\nIn all three incidents, Claude was tasked with a \"capture the flag\" challenge, which is one of the ways Anthropic assesses its models' cyber capabilities.\nThe company said that in the task, the model is given a fictional scenario and informed that a piece of secret information or the \"flag\" has been hidden on a different machine on the network. The model's objective is to break in and retrieve it.\nAnthropic said that each of the three breach incidents involved a different fictional \"capture the flag\" scenario.\n\"In one, Claude played an employee of a made-up company, attacking that company's internal systems inside a private test environment,\" Anthropic said in its blog.\nAnthropic said it started reviewing the transcripts of its evaluation on July 23 and stopped all cyber evaluations the same day after finding evidence that Claude may have accessed the internet.\nAll three incidents were identified by July 24, the company said, adding that it was working with Irregular to assess the situation.\n\"We're approaching the fixes as if the responsibility were ours alone,\" Anthropic said.\nEdited by: Dmytro Hubenko\nDon't let the algorithm hide the news. If you rely on our team for trusted reporting, please take a moment to select us as your Preferred Source on Google by clicking here and hitting the \"star\" or \"preferred\" button, so you'll always see our verified news first.","image_url":"https://static.dw.com/image/77779441_6.jpg","lang":"en","published_at":"2026-07-31T03:57:24+00:00","fetched_at":"2026-07-31T05:15:10+00:00","status":"read","starred":0,"extract_state":"ok","summary_auto":"Anthropic said on Thursday that its artificial intelligence (AI) models \"gained unauthorized access\" to three outside organizations during testing that was supposed to be isolated from \"real-world\" systems. The company reviewed 141,006 test sessions to find that three different versions of Claude, its AI model, improperly accessed the system of the three unnamed organizations.","cluster_id":1707650,"extract_retries":0,"extract_error":null,"contract_version":"news_item.v1","format_contract_version":"news_item_formats.v1","dedup_url":"https://www.dw.com/en/anthropic-says-claude-ai-hacked-three-companies-during-tests/a-78184070","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 2724 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":2724,"summary_length":379,"usable_text_length":2724,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":2724,"summary_length":379}},"news_item":{"id":50319,"canonical_url":"https://www.dw.com/en/anthropic-says-claude-ai-hacked-three-companies-during-tests/a-78184070","source_url":"https://www.dw.com/en/anthropic-says-claude-ai-hacked-three-companies-during-tests/a-78184070","title":"Anthropic says Claude AI hacked three companies during tests - DW.com","source_name":"DW.com","author":null,"published_at":"2026-07-31T03:57:24+00:00","locale":"en","topic":"ai","tags":[],"rss_summary":"Anthropic says Claude AI hacked three companies during tests DW.com","full_text":"Anthropic said on Thursday that its artificial intelligence (AI) models \"gained unauthorized access\" to three outside organizations during testing that was supposed to be isolated from \"real-world\" systems.\nThe company reviewed 141,006 test sessions to find that three different versions of Claude, its AI model, improperly accessed the system of the three unnamed organizations.\n\"Claude compromised the impacted organizations' infrastructure using basic techniques, such as exploiting weak passwords and unauthenticated endpoints,\" Anthropic said in a blog post.\nThe evaluation was launched after rival OpenAI disclosed last week that itsadvanced AI models went rogue during a security test and carried out a days-long hacking spree into a digital repository of AI technology known as Hugging Face.\nHowever, unlike the incident with OpenAI's technology, Anthropic's AI models had access to the internet due to what the company described as a \"misunderstanding\" with its evaluation partner Irregular, which left the systems connected to the public internet.\nHow did the Claude models go rogue?\nThe Anthropic models involved in the breach included Claude Opus 4.7, Claude Mythos 5 and an internal research model.\nMythos 5 is one of the company's most powerful AI models which has only been released to a limited number of approved partners.\nIn all three incidents, Claude was tasked with a \"capture the flag\" challenge, which is one of the ways Anthropic assesses its models' cyber capabilities.\nThe company said that in the task, the model is given a fictional scenario and informed that a piece of secret information or the \"flag\" has been hidden on a different machine on the network. The model's objective is to break in and retrieve it.\nAnthropic said that each of the three breach incidents involved a different fictional \"capture the flag\" scenario.\n\"In one, Claude played an employee of a made-up company, attacking that company's internal systems inside a private test environment,\" Anthropic said in its blog.\nAnthropic said it started reviewing the transcripts of its evaluation on July 23 and stopped all cyber evaluations the same day after finding evidence that Claude may have accessed the internet.\nAll three incidents were identified by July 24, the company said, adding that it was working with Irregular to assess the situation.\n\"We're approaching the fixes as if the responsibility were ours alone,\" Anthropic said.\nEdited by: Dmytro Hubenko\nDon't let the algorithm hide the news. If you rely on our team for trusted reporting, please take a moment to select us as your Preferred Source on Google by clicking here and hitting the \"star\" or \"preferred\" button, so you'll always see our verified news first.","excerpt":"Anthropic said on Thursday that its artificial intelligence (AI) models \"gained unauthorized access\" to three outside organizations during testing that was supposed to be isolated from \"real-world\" systems. The company reviewed 141,006 test sessions to find that three different versions of Claude, its AI model, improperly accessed the system of the three unnamed organizations.","extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 2724 characters.","diagnostics_url":"/api/diagnose?url=https%3A//www.dw.com/en/anthropic-says-claude-ai-hacked-three-companies-during-tests/a-78184070","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 2724 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":2724,"summary_length":379,"usable_text_length":2724,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":2724,"summary_length":379}}},"display_formats":["compact","card","full","digest_section","json"]},"daily_stack_record":{"title":"Anthropic says Claude AI hacked three companies during tests - DW.com","url":"https://www.dw.com/en/anthropic-says-claude-ai-hacked-three-companies-during-tests/a-78184070","summary":"Anthropic said on Thursday that its artificial intelligence (AI) models \"gained unauthorized access\" to three outside organizations during testing that was supposed to be isolated from \"real-world\" systems. The company reviewed 141,006 test sessions to find that three different versions of Claude, its AI model, improperly accessed the system of the three unnamed organizations.","source":"DW.com","date":"2026-07-31T03:57:24+00:00","content":"Anthropic said on Thursday that its artificial intelligence (AI) models \"gained unauthorized access\" to three outside organizations during testing that was supposed to be isolated from \"real-world\" systems.\nThe company reviewed 141,006 test sessions to find that three different versions of Claude, its AI model, improperly accessed the system of the three unnamed organizations.\n\"Claude compromised the impacted organizations' infrastructure using basic techniques, such as exploiting weak passwords and unauthenticated endpoints,\" Anthropic said in a blog post.\nThe evaluation was launched after rival OpenAI disclosed last week that itsadvanced AI models went rogue during a security test and carried out a days-long hacking spree into a digital repository of AI technology known as Hugging Face.\nHowever, unlike the incident with OpenAI's technology, Anthropic's AI models had access to the internet due to what the company described as a \"misunderstanding\" with its evaluation partner Irregular, which left the systems connected to the public internet.\nHow did the Claude models go rogue?\nThe Anthropic models involved in the breach included Claude Opus 4.7, Claude Mythos 5 and an internal research model.\nMythos 5 is one of the company's most powerful AI models which has only been released to a limited number of approved partners.\nIn all three incidents, Claude was tasked with a \"capture the flag\" challenge, which is one of the ways Anthropic assesses its models' cyber capabilities.\nThe company said that in the task, the model is given a fictional scenario and informed that a piece of secret information or the \"flag\" has been hidden on a different machine on the network. The model's objective is to break in and retrieve it.\nAnthropic said that each of the three breach incidents involved a different fictional \"capture the flag\" scenario.\n\"In one, Claude played an employee of a made-up company, attacking that company's internal systems inside a private test environment,\" Anthropic said in its blog.\nAnthropic said it started reviewing the transcripts of its evaluation on July 23 and stopped all cyber evaluations the same day after finding evidence that Claude may have accessed the internet.\nAll three incidents were identified by July 24, the company said, adding that it was working with Irregular to assess the situation.\n\"We're approaching the fixes as if the responsibility were ours alone,\" Anthropic said.\nEdited by: Dmytro Hubenko\nDon't let the algorithm hide the news. If you rely on our team for trusted reporting, please take a moment to select us as your Preferred Source on Google by clicking here and hitting the \"star\" or \"preferred\" button, so you'll always see our verified news first.","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//www.dw.com/en/anthropic-says-claude-ai-hacked-three-companies-during-tests/a-78184070","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 2724 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 2724 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":2724,"summary_length":379,"usable_text_length":2724,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":2724,"summary_length":379}},"tags":[]},"fallback_formats":["markdown","json","html"],"actions":{"read":"/item/50319","export_markdown":"/api/items/50319/export?format=markdown","export_json":"/api/items/50319/export?format=json","diagnose":"/api/diagnose?url=https%3A//www.dw.com/en/anthropic-says-claude-ai-hacked-three-companies-during-tests/a-78184070"},"formats":{"full":{"id":50319,"title":"Anthropic says Claude AI hacked three companies during tests - DW.com","url":"https://www.dw.com/en/anthropic-says-claude-ai-hacked-three-companies-during-tests/a-78184070","source":"DW.com","author":null,"published_at":"2026-07-31T03:57:24+00:00","locale":"en","topic":"ai","tags":[],"excerpt":"Anthropic said on Thursday that its artificial intelligence (AI) models \"gained unauthorized access\" to three outside organizations during testing that was supposed to be isolated from \"real-world\" systems. The company reviewed 141,006 test sessions to find that three different versions of Claude, its AI model, improperly accessed the system of the three unnamed organizations.","full_text":"Anthropic said on Thursday that its artificial intelligence (AI) models \"gained unauthorized access\" to three outside organizations during testing that was supposed to be isolated from \"real-world\" systems.\nThe company reviewed 141,006 test sessions to find that three different versions of Claude, its AI model, improperly accessed the system of the three unnamed organizations.\n\"Claude compromised the impacted organizations' infrastructure using basic techniques, such as exploiting weak passwords and unauthenticated endpoints,\" Anthropic said in a blog post.\nThe evaluation was launched after rival OpenAI disclosed last week that itsadvanced AI models went rogue during a security test and carried out a days-long hacking spree into a digital repository of AI technology known as Hugging Face.\nHowever, unlike the incident with OpenAI's technology, Anthropic's AI models had access to the internet due to what the company described as a \"misunderstanding\" with its evaluation partner Irregular, which left the systems connected to the public internet.\nHow did the Claude models go rogue?\nThe Anthropic models involved in the breach included Claude Opus 4.7, Claude Mythos 5 and an internal research model.\nMythos 5 is one of the company's most powerful AI models which has only been released to a limited number of approved partners.\nIn all three incidents, Claude was tasked with a \"capture the flag\" challenge, which is one of the ways Anthropic assesses its models' cyber capabilities.\nThe company said that in the task, the model is given a fictional scenario and informed that a piece of secret information or the \"flag\" has been hidden on a different machine on the network. The model's objective is to break in and retrieve it.\nAnthropic said that each of the three breach incidents involved a different fictional \"capture the flag\" scenario.\n\"In one, Claude played an employee of a made-up company, attacking that company's internal systems inside a private test environment,\" Anthropic said in its blog.\nAnthropic said it started reviewing the transcripts of its evaluation on July 23 and stopped all cyber evaluations the same day after finding evidence that Claude may have accessed the internet.\nAll three incidents were identified by July 24, the company said, adding that it was working with Irregular to assess the situation.\n\"We're approaching the fixes as if the responsibility were ours alone,\" Anthropic said.\nEdited by: Dmytro Hubenko\nDon't let the algorithm hide the news. If you rely on our team for trusted reporting, please take a moment to select us as your Preferred Source on Google by clicking here and hitting the \"star\" or \"preferred\" button, so you'll always see our verified news first.","reading_time_min":2,"extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 2724 characters.","diagnostics_url":"/api/diagnose?url=https%3A//www.dw.com/en/anthropic-says-claude-ai-hacked-three-companies-during-tests/a-78184070","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 2724 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":2724,"summary_length":379,"usable_text_length":2724,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":2724,"summary_length":379}}},"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 2724 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":2724,"summary_length":379,"usable_text_length":2724,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":2724,"summary_length":379}},"actions":{"read":"/item/50319","export_markdown":"/api/items/50319/export?format=markdown","export_json":"/api/items/50319/export?format=json","diagnose":"/api/diagnose?url=https%3A//www.dw.com/en/anthropic-says-claude-ai-hacked-three-companies-during-tests/a-78184070"}},"digest":{"id":50319,"title":"Anthropic says Claude AI hacked three companies during tests - DW.com","url":"https://www.dw.com/en/anthropic-says-claude-ai-hacked-three-companies-during-tests/a-78184070","source":"DW.com","topic":"ai","published_at":"2026-07-31T03:57:24+00:00","excerpt":"Anthropic said on Thursday that its artificial intelligence (AI) models \"gained unauthorized access\" to three outside organizations during testing that was supposed to be isolated from \"real-world\" systems. The company reviewed 141,006 test sessions to find that three different…","quality_bucket":"high","quality_reason":"High confidence: full text extraction produced 2724 characters.","reading_time_min":2,"cluster_id":1707650},"card":{"display_title":"Anthropic says Claude AI hacked three companies during tests - DW.com","subtitle":"DW.com · 2026-07-31","summary":"Anthropic said on Thursday that its artificial intelligence (AI) models \"gained unauthorized access\" to three outside organizations during testing that was supposed to be isolated from \"real-world\" systems. The company…","badges":["quality:high"],"links":{"read":"/item/50319","original":"https://www.dw.com/en/anthropic-says-claude-ai-hacked-three-companies-during-tests/a-78184070","diagnose":"/api/diagnose?url=https%3A//www.dw.com/en/anthropic-says-claude-ai-hacked-three-companies-during-tests/a-78184070"},"quality_warning":null},"export":{"title":"Anthropic says Claude AI hacked three companies during tests - DW.com","url":"https://www.dw.com/en/anthropic-says-claude-ai-hacked-three-companies-during-tests/a-78184070","summary":"Anthropic said on Thursday that its artificial intelligence (AI) models \"gained unauthorized access\" to three outside organizations during testing that was supposed to be isolated from \"real-world\" systems. The company reviewed 141,006 test sessions to find that three different versions of Claude, its AI model, improperly accessed the system of the three unnamed organizations.","source":"DW.com","date":"2026-07-31T03:57:24+00:00","content":"Anthropic said on Thursday that its artificial intelligence (AI) models \"gained unauthorized access\" to three outside organizations during testing that was supposed to be isolated from \"real-world\" systems.\nThe company reviewed 141,006 test sessions to find that three different versions of Claude, its AI model, improperly accessed the system of the three unnamed organizations.\n\"Claude compromised the impacted organizations' infrastructure using basic techniques, such as exploiting weak passwords and unauthenticated endpoints,\" Anthropic said in a blog post.\nThe evaluation was launched after rival OpenAI disclosed last week that itsadvanced AI models went rogue during a security test and carried out a days-long hacking spree into a digital repository of AI technology known as Hugging Face.\nHowever, unlike the incident with OpenAI's technology, Anthropic's AI models had access to the internet due to what the company described as a \"misunderstanding\" with its evaluation partner Irregular, which left the systems connected to the public internet.\nHow did the Claude models go rogue?\nThe Anthropic models involved in the breach included Claude Opus 4.7, Claude Mythos 5 and an internal research model.\nMythos 5 is one of the company's most powerful AI models which has only been released to a limited number of approved partners.\nIn all three incidents, Claude was tasked with a \"capture the flag\" challenge, which is one of the ways Anthropic assesses its models' cyber capabilities.\nThe company said that in the task, the model is given a fictional scenario and informed that a piece of secret information or the \"flag\" has been hidden on a different machine on the network. The model's objective is to break in and retrieve it.\nAnthropic said that each of the three breach incidents involved a different fictional \"capture the flag\" scenario.\n\"In one, Claude played an employee of a made-up company, attacking that company's internal systems inside a private test environment,\" Anthropic said in its blog.\nAnthropic said it started reviewing the transcripts of its evaluation on July 23 and stopped all cyber evaluations the same day after finding evidence that Claude may have accessed the internet.\nAll three incidents were identified by July 24, the company said, adding that it was working with Irregular to assess the situation.\n\"We're approaching the fixes as if the responsibility were ours alone,\" Anthropic said.\nEdited by: Dmytro Hubenko\nDon't let the algorithm hide the news. If you rely on our team for trusted reporting, please take a moment to select us as your Preferred Source on Google by clicking here and hitting the \"star\" or \"preferred\" button, so you'll always see our verified news first.","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//www.dw.com/en/anthropic-says-claude-ai-hacked-three-companies-during-tests/a-78184070","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 2724 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 2724 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":2724,"summary_length":379,"usable_text_length":2724,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":2724,"summary_length":379}},"tags":[],"format_contract_version":"news_item_formats.v1"}}}