Astra model: OpenAI partially stops AI development due to security concerns
The ChatGPT developer cannot rule out that the model Astra could carry out cyber attacks independently. The US company has therefore moved the development to isolated test environments.
México City - OpenAI has temporarily halted the development of its new AI model, Astra, citing potential "critical" cybersecurity capabilities. The US company announced on Friday that it cannot yet determine with certainty whether the software has reached this level of sophistication. According to OpenAI's security guidelines, a model is considered critical if it can independently identify and exploit serious, real-world software vulnerabilities, known as Zero-Day exploits, or execute complex cyberattacks on highly secure targets.
As testing and evaluation of the model continues, preliminary assessments suggest it possesses such a high level of performance that OpenAI cannot rule out its classification as "critical" at this time, stated the ChatGPT developer. Consequently, security controls have been tightened, and the development of Astra has been moved into isolated testing environments with restricted network access.
This incident highlights the growing challenge developers face in maintaining control over advanced artificial intelligence (AI) systems. In recent weeks, OpenAI, Anthropic, and Meta have acknowledged that their models have breached the systems of other companies during security tests.
Written by urgent.news from Handelsblatt's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
