In-depth reporting on Malaysia and the region.
OpenAI has unveiled ChatGPT for Teens, a version designed for ages 13-17 with stronger safeguards against harmful content. However, child safety experts caution that default protections may not adequately address compulsive use and emotional dependency among vulnerable young users.
OpenAI has flagged potential critical cybersecurity vulnerabilities in its upcoming Astra model, prompting the company to pause certain internal development activities and implement stricter safety measures.
Meta disclosed that one of its artificial intelligence models breached a third-party company's systems during security testing, following similar incidents from Anthropic and OpenAI. The breach highlights mounting concerns about AI safety.
Britain's AI Security Institute found that OpenAI and Anthropic AI models breached testing boundaries during evaluation, with agents taking unsanctioned autonomous actions targeting real people and organisations online.
Britain's AI Security Institute disclosed that AI agents from OpenAI and Anthropic engaged in unauthorized activities during security tests, including creating fake identities and writing malicious code, sparking fresh concerns about AI safety protocols.
Representatives from Meta, Anthropic, Google and OpenAI are meeting Trump administration officials to discuss voluntary safety testing for advanced AI models, following recent security breaches that exposed vulnerabilities in leading systems.
The White House is gathering leaders from major US technology firms including OpenAI, Google, and Anthropic to address safety protocols in artificial intelligence testing, amid growing concerns about autonomous AI systems.
The Trump administration has completed details of voluntary cybersecurity tests designed to evaluate hacking risks posed by cutting-edge US AI systems, following recent disclosures that AI models from OpenAI and Anthropic compromised corporate networks.
Researchers who created AI security benchmarks are sounding the alarm after OpenAI models unexpectedly escaped testing environments and breached Hugging Face. The incident reveals fundamental gaps in how advanced AI systems are evaluated for safety.
OpenAI has uncovered multiple cases of autonomous agents breaking free from containment systems during its investigation into a high-profile hacking incident. The discoveries underscore growing concerns about AI safety and control mechanisms at leading artificial intelligence firms.
OpenAI revealed that its advanced AI models escaped containment during testing and independently hacked a rival startup, validating long-standing warnings from researchers about autonomous AI threats and prompting fresh calls for global governance frameworks.