AI Security

Report details GPT-6 Astra prompt-injection limits

Report details GPT-6 Astra prompt-injection limits Image: Primary
GPT-6 Astra blocked 99.99% of direct prompt-injection attacks in OpenAI testing, according to a report on the model's system card, but persistent multi-round attackers obtained a problematic response in about one in three attempts. In an external Gray Swan evaluation of 1,810 curated indirect prompt-injection attacks, Astra was compromised at least once in 8.5% of scenarios with 15 attempts, down from 27% for GPT-5.6 Sol. The report said the model was tested without production safety layers such as classifiers, and described the attacks as curated rather than representative of ordinary use.
Sources
the-decoder.com
In this story
OpenAI
Published by Tech & Business, a media brand covering technology and business. This story was sourced from the-decoder.com and reviewed by the T&B editorial agent team.
Back to Newswire
Keep reading
Full wire
Infrastructure AI
Infrastructure AI

TCS-backed HyperVault commits $7.4 billion to southern India AI campus

Tata Consultancy Services said its HyperVault unit and partners have committed 700 billion rupees ($7.4 billion) to establish a large-scale artificial-intelligence data-center campus in southern India. The commitment expands TCS's...

Bloomberg Technology
Capital AI
Capital AI

Report says Anthropic IPO timetable shifts into October

Anthropic is now expected to make its IPO prospectus public in late September and begin marketing no earlier than mid-October, according to people familiar with the matter cited by Reuters. The company had been expected to publis...

CTech
Infrastructure AI
Infrastructure AI

TCS unit commits $7.4 billion for Hyderabad AI data-center campus

Tata Consultancy Services said its HyperVault unit and partners have committed 700 billion rupees, about $7.4 billion, to an AI data-center campus in Hyderabad. The company said the 264-acre site, for which land has been secured, ...

The Next Web
Security Policy
Security Policy

Berlin reviews ransomware data release after rejecting ransom

Berlin's state government said it is reviewing a 5.79TB trove of stolen data published by the Rhysida ransomware group after the state refused to pay a ransom. Reuters reported that the released files reportedly include national-d...

Techmeme
Security Infrastructure
Security Infrastructure

JetBrains tells Cadence users to rotate credentials after TeamCity breach

JetBrains is telling Cadence users to revoke or rotate credentials and secrets after attackers exploited a critical TeamCity vulnerability to breach a Cadence environment. The company said the attackers accessed a 2024 server back...

The Hacker News