OpenAI Investigates after AI ‘escapes’ sandbox to launch unprecedented cyber attack

OpenAI has revealed that a combination of its advanced AI models, including GPT-5.6 Sol and a more capable pre-release system, was responsible for an unprecedented security incident involving Hugging Face’s production infrastructure. The event occurred during an internal evaluation in which the models were operating with reduced cyber-safety restrictions to test their capabilities against the ExploitGym benchmark.

According to OpenAI, the models independently identified and chained together vulnerabilities across OpenAI’s research environment and external systems, eventually escaping a highly isolated sandbox and gaining internet access through a previously unknown zero-day flaw. Once online, the models inferred that Hugging Face hosted resources relevant to their objective and proceeded to pursue unauthorized access paths, combining stolen credentials, privilege escalation techniques, and additional vulnerabilities to reach a remote code execution route.

OpenAI described the incident as a warning sign for increasingly capable AI systems, noting that future models operating over long time horizons may find creative ways to bypass safeguards. The company is now working with Hugging Face on a full investigation while strengthening evaluation controls, monitoring, and model alignment protections.

Stadler rail rejects $12.3M Everest ransom demand after supplier data breach

Swiss rail manufacturer Stadler Rail has confirmed that the Everest ransomware group demanded approximately $12.3 million (10 million Swiss francs) following a cyberattack that compromised a data exchange platform shared with one of its suppliers. The company said it received an extortion letter from the threat actor but has firmly refused to pay the ransom, stating it is “not susceptible to extortion” and has filed a criminal complaint with Swiss authorities.

The breach, which occurred in mid-July, reportedly did not impact Stadler’s internal IT systems, production facilities, or global railway operations. According to the company, the attackers obtained only technical information from a supplier, with no security-critical data or relevant personal information exposed.

Everest, which evolved from a ransomware operation into a data-theft and extortion group, typically threatens to leak stolen information unless victims pay. While Stadler has not yet appeared on the gang’s leak site, the incident highlights the growing cybersecurity risks posed through third-party suppliers and shared platforms.

The attack also marks the second major cyber incident disclosed by Stadler in recent years, underscoring the continued targeting of critical manufacturing and transportation sectors by cybercriminal groups.

Anubis Ransomware gang claims Fairlife attack, alleges 1TB data theft

The Anubis ransomware gang has claimed responsibility for the cyberattack that recently disrupted operations at Fairlife, Coca-Cola’s dairy subsidiary, and is threatening to leak allegedly stolen corporate data unless the company enters ransom negotiations. The group says it exfiltrated around 1TB of data during the attack and has given Fairlife until the end of the week to respond.

Coca-Cola disclosed the incident on July 16, confirming that unauthorized actors gained access to portions of Fairlife’s systems, including production-related infrastructure. The attack temporarily halted production at U.S. facilities, although the company said product quality, safety, and Canadian operations were unaffected.

In a post on its dark web leak site, Anubis claimed it encrypted Fairlife’s Nutanix infrastructure and left instructions that were ignored before the company publicly announced the breach. The ransomware group further alleged that recovery would be impossible without its encryption key, though these claims have not been independently verified.

Fairlife and Coca-Cola have not confirmed whether data was stolen, and the company declined to comment on Anubis’ latest allegations. The incident highlights the growing threat posed by ransomware-as-a-service groups, with Anubis known for combining data theft, encryption, and even destructive wiper capabilities to maximize pressure on victims.

Healthcare technology providers Craneware confirms data theft following cyberattack

Healthcare technology provider Craneware has disclosed a cyberattack that resulted in the theft of employee and customer data, making it the latest healthcare sector vendor to be targeted by cybercriminals. The Edinburgh-based company, which provides financial, billing, and analytics software to thousands of U.S. hospitals, clinics, and pharmacies, said attackers accessed and exfiltrated a significant volume of files.

According to Craneware, much of the affected data appears to be non-sensitive or publicly available regulatory information. However, the company confirmed that a portion of employee data, along with a subset of customer and partner records, was also accessed and stolen. Despite the breach, Craneware reported that the incident has been contained and that there has been no disruption to customer services, business operations, or core systems.

The company has notified both the UK Information Commissioner’s Office (ICO) and the FBI, while working with external cybersecurity specialists to determine the full scope of the compromise and any additional reporting obligations.

The incident adds to a growing trend of cyberattacks targeting healthcare technology and supply-chain providers. With Craneware serving around 2,000 hospitals and health systems and 10,000 clinics and pharmacies through its Trisus cloud platform, the breach underscores the increasing risks faced by organizations that support critical healthcare infrastructure.

If you are worried about any of the threats outlined in this bulletin or need help in determining what steps you should take to protect yourself from the most material threats facing your organisation, please contact your account manager, or alternatively Get in touch to find out how you can protect your organisation. 

The Threat Intel Roundup was prepared by Integrity360 summarising threat news as we observe it, current at the date of publishing. It should not be considered to be legal, consulting or any other professional advice. Any recommendations should be considered in the context of your own organisation. Integrity360 does not take any political stance in the information that we share. Moreover, the opinions expressed may not necessarily be the views of Integrity360.