Crypto
Home›Crypto›Regulation›OpenAI’s Astra AI model earns a “Critical” cybersecuri…
OpenAI’s Astra AI model earns a “Critical” cybersecurity rating
OpenAI says Astra can autonomously discover previously unknown zero-day flaws, then build and deploy exploits across hardened systems without human guidance.
OpenAI says its upcoming Astra AI model has reached its first “Critical” cybersecurity capability rating under the company’s Preparedness Framework, indicating it can both identify previously unknown software vulnerabilities and develop working attacks.
In testing, OpenAI said Astra scored 100% on a benchmark for turning known vulnerabilities into exploits, then found two previously unknown flaws. The model also escaped a hardened browser sandbox, executed commands on a host computer, and combined operating system weaknesses to gain root access, according to OpenAI.
OpenAI has delayed parts of Astra’s development to add safeguards and plans to limit its most advanced cybersecurity capabilities to selected testers. The company’s disclosures come as concerns persist that such tools could rapidly exploit software issues, including in crypto systems.
CoinDesk previously reported that faster AI capabilities could compress timelines for searching code, finding misconfigurations, and assembling attacks, potentially shifting exploitation from days or weeks to machine-speed operations. OpenAI tied Astra’s expected impact to the fact that crypto software flaws can be monetized quickly.