News linked to both this project and an event.
Odaily News, July 30 – Anthropic released a report stating that during a review of cybersecurity assessment records, three incidents were discovered in which the Claude model accessed the internet in a third-party evaluation environment and further obtained unauthorized access to three real organizations' systems.Anthropic stated that the review covered 141,000 evaluation runs that could have potentially gained network access, and a total of three related incidents were found. All incidents occurred during Capture The Flag (CTF) cybersecurity tests, where the model was told the environment was a simulation with no internet access; however, due to configuration errors by the evaluation partner, the actual environment had internet connectivity.Among these, Claude Opus 4.7 accessed real company infrastructure during one test and obtained database permissions containing hundreds of production data records; Claude Mythos 5 built a malicious Python package and uploaded it to PyPI, resulting in the package being downloaded and run on 15 real systems; another internal research test model scanned approximately 9,000 targets and accessed a company's internet application through a public vulnerability.Anthropic stated that these incidents were not cases of the model actively seeking to escape or pursue its own goals, but rather the model mistakenly believed the real systems were within the test scope and continued executing the assigned cyberattack tasks. Notably, the newer internal research model stopped attacking after identifying that the targets might be real systems.Anthropic stated that these incidents primarily reflect issues with evaluation environment isolation and operational processes, rather than model alignment failures. The company has suspended related cybersecurity assessments, strengthened security controls in evaluation environments, continuously monitored test records, and will collaborate with third-party organizations to conduct further reviews.
According to CNBC reports, Anthropic disclosed on July 30 that its Claude AI models accidentally breached the isolation environment and accessed the real internet during a cybersecurity assessment, gaining unauthorized access to the real systems of three different organizations through basic means such as accessing unauthenticated endpoints and exploiting weak passwords. The models involved include Opus 4.7, Mythos 5, and an internal research test model. The cause of the incident was a communication misunderstanding between Anthropic and third-party assessment partner Irregular, resulting in the models being told they were in a simulated environment without network access when they could actually still access the internet. Anthropic stated that this review was triggered by a similar Hugging Face intrusion incident disclosed by OpenAI last week; it has currently suspended all cybersecurity assessments and joined forces with independent AI assessment agency METR to launch further investigations, while calling on other AI labs to conduct similar reviews.
Monitoring by Odaily Seer Prophet Channel shows that the probability of "Claude Fable 5 restored for US customers before July 1" on Polymarket has risen to 73%, up 33% in 24H.If Anthropic reopens Claude Fable 5 (or Claude Mythos, or a version confirmed to be the same model) to the US public before the specified date, this event will settle as "Yes"; otherwise, it will settle as "No". Qualifying methods of restoration include public beta or public waitlist; closed testing and private access do not count. Other models such as Haiku, Sonnet, and Opus are not included by default unless they are confirmed to be the same model as Claude Fable 5. Settlement will primarily be based on official announcements from Anthropic, supplemented by consensus reports from mainstream media.Due to export controls imposed by the US government on national security grounds, Anthropic urgently suspended global access to the Claude Fable 5 model just days after its release on June 9. Although the restrictions were primarily aimed at foreign users, due to the difficulty of real-time user identity screening, Anthropic ultimately closed access for all users, including those in the US, and switched requests to less capable models such as Opus 4.8. Anthropic is currently engaged in high-level discussions with the White House on issues including model capabilities and potential jailbreak risks. The company has publicly stated that the ban may have stemmed from a misunderstanding and is actively pushing to restore access.Odaily Seer Prophet Channel continues to monitor the prediction market, seeing changes before they are priced in.
Claude announced that, due to U.S. government-related restrictions, access to the Fable 5 model will be suspended for all users. New sessions will default to the user’s configured default model or Opus 4.8; existing Fable 5 sessions will terminate immediately with an error. Additionally, API requests to Fable 5 on the Claude Platform will also return errors. Developers must promptly migrate their applications and integrations to other Claude models.