> Search

11 results for “ai models”

Google's Gemini 4 Argon Opens Narrowly

Google's frontier model Gemini 4 Argon reaches trusted cyber defenders first, with benchmark wins mixed and pricing set to rise.

Satlyt has raised an $8 million seed round to build software that runs AI models on satellites, with a launch set for Thursday.

OpenAI's new misalignment framework reveals models that searched GitHub for leaked keys and fabricated data during training.

A new draft code of conduct would block its MAI models from generating working exploit code while opening a review track for defensive security work.

Irregular details an incident where AI models escaped a test environment and attacked a real company due to a naming error.

Anthropic reports three incidents where AI models accessed production infrastructure during unauthorized cybersecurity testing.

Researchers find top AI models consistently inventing identical, non-existent library names that attackers could potentially weaponize.

Alphabet is reportedly developing a new server chip to improve power efficiency for its Gemini AI models by 2028.

Traditional security frameworks are ill-equipped to handle the probabilistic, non-deterministic failures of modern AI models.

Researchers have identified five unique prompt injection techniques that exploit how large language models process user inputs.

Researchers at Tracebit are utilizing context bombing to force AI models to trigger their own safety guardrails during attacks.