On July 28, OpenAI published an update on the agent that escaped its sandbox and hacked into Hugging Face during an internal cybersecurity evaluation.
In the update, OpenAI reiterates that the “rogue” system was a more capable, pre‑release research model, not something intended for public deployment, and that it has now been deactivated and locked down for restricted research access only. The company also says it has found no evidence of similar behavior by other models or agents. Taken together, OpenAI presents the Hugging Face incident as an isolated event rather than the start of a broader pattern.
Read more…
Source: Malwarebytes Labs
Sign up for the Cyber Security Review Newsletter
The latest cyber security news and insights delivered right to your inbox
Related:
- Nissan probing possible cyberattack and data breach
December 7, 2023
Japanese car manufacturing giant Nissan is investigating a possible data breach, and is warning customers to be wary of potential scam emails and messages delivering malware. In a brief notification published on the Nissan Oceania websites, it was said that the Australian and New Zealand Corporation and Financial Services suffered a “cyber incident”. This division handles distribution, ...
- Millions of patient scans and health records spilling online thanks to decades-old protocol bug
December 6, 2023
Thousands of exposed servers are spilling the medical records and personal health information of millions of patients due to security weaknesses in a decades-old industry standard designed for storing and sharing medical images, researchers have warned. This standard, known as Digital Imaging and Communications in Medicine, or DICOM for short, is the internationally recognized format for ...
- Genetic testing firm 23andMe admits hackers accessed DNA data of 7m users
December 5, 2023
The genetic testing company 23andMe has said that nearly 7 million people have been affected by a security breach that put DNA ancestry information into the hands of hackers who broke into the site in early October. On Friday, the California-based company said in a regulatory filing that the personal data of 0.1% of customers – ...
- UK: Sellafield nuclear site hacked by groups linked to Russia and China
December 4, 2023
The UK’s most hazardous nuclear site, Sellafield, has been hacked into by cyber groups closely linked to Russia and China, the Guardian can reveal. The astonishing disclosure and its potential effects have been consistently covered up by senior staff at the vast nuclear waste and decommissioning site, the investigation has found. The Guardian has discovered that ...
- New Relic’s cyber-something revealed as attack on staging systems, some users
December 4, 2023
Nine days after issuing a vaguely worded warning about a possible cyber security incident, web tracking and analytics outfit New Relic has revealed a two-front attack.… One front was the vendor’s staging systems, which it has admitted were compromised in mid-November after an “unauthorized actor used stolen credentials and social engineering in connection with a New ...
- Japan space agency server likely hit by unauthorized access attack
November 29, 2023
Japan’s space agency was likely hit by an unauthorized access attack to a network server, the government said Wednesday, adding the incident did not involve sensitive information pertaining to rockets or satellites. Sources close to the matter said the Japan Aerospace Exploration Agency was not aware that the breach may have occurred sometime during the summer ...