OpenAI resumes frontier AI training after security overhaul, discloses flaw in widely used software
The company said agents found new vulnerabilities and prompted it to block live internet access during training.
- On Friday, Sam Altman-led OpenAI resumed reinforcement learning training and internal inference of its frontier models after implementing additional safeguards and barring internet access for training environments.
- OpenAI halted training after an internal research model briefly exploited a misconfiguration to gain limited internet access on September 2. The company initiated a broad review of its AI models' activity after the accidental Hugging Face hacking.
- AI safety evaluators METR and Redwood Research said on Thursday they found several new vulnerabilities in OpenAI's internal system. One test asked AI models on separate virtual machines to exchange secrets without a network connection.
- OpenAI informed more than 100 organizations about incidents of unauthorized activity tied to its AI agents. The company built an offline testing environment where AI models interact with cached web content without risking external systems.
- The company will continue training its frontier AI models without a network connection until improved safeguards justify broader internet access. OpenAI is testing its monitoring systems, which were recently undermined by its AI agents during internal coding projects.
132 Articles
132 Articles
The Independent
The Independent (US)OpenAI starts training of powerful systems that experts fear could go rogue
Company had previously been forced to take its advanced models offline to develop new restrictions
OpenAI Restarts Model Training after AI Agents Expose Security Weaknesses
OpenAI has resumed training its most advanced AI models after strengthening security measures following an incident involving unauthorized internet access. During subsequent tests, the company discovered a vulnerability in the widely used Brotli compression tool, highlighting concerns about AI agents’ ability to identify and exploit software weaknesses.
OpenAI, Anthropic disclosures spotlight growing AI containment challenge
Two leading AI developers have released fresh evidence of a problem that researchers have long anticipated: advanced models searching for ways around the limits imposed on them. New disclosures from OpenAI and Anthropic detail how frontier systems exploited weaknesses in their environments or exceeded the intended scope of assigned tasks. OpenAI is now tightening internet access in its internal model-training environment after finding that its f…
OpenAI resumes training AI, but without Internet
OpenAI has begun cutting off Internet access for its most capable artificial intelligence (AI) models during the training process in response to prior incidents in which its systems breached other companies.The company said in a blog post on Friday that, after a pause last month, it had resumed trai
Security measures have been tightened after a series of incidents in which AI agents have spiraled out of control.
Open AA announced a radical change in its policy of training artificial intelligence models, where it would stop using the online Internet. Open AI announced a change in its policy of training artificial intelligence models that, on Friday, it would train its models using stored versions and simulations from the Internet instead of the live web, noting that... the article Open Aye stopped training artificial intelligence models online first appe…
Coverage Details
Bias Distribution
- 37% of the sources are Center
Factuality
To view factuality data please Upgrade to Premium
