NVIDIA maintains strict no-backdoor policy for its GPUs
NVIDIA asserts its GPUs come free of backdoors, kill switches, or spyware, emphasizing user control for secure Artificial Intelligence and data center operations.
Protocols aim to help artificial intelligence agents manage complex digital tasks
New protocols from Anthropic and Google aim to help Artificial Intelligence agents navigate and interact with the diverse elements of our digital lives, tackling crucial security, openness, and efficiency challenges.
Breakthroughs in artificial intelligence: security, agent communication, math, and video generation
Discover four transformative innovations redefining the Artificial Intelligence landscape, from red teaming security to seamless agent collaboration, advanced reasoning, and next-gen video creation.
Benchmarking large language model vulnerability to insecure code via few-shot inversion
Researchers introduce a novel few-shot inversion method to systematically evaluate how large language models generate insecure code, creating the first open benchmark for code security risks in Artificial Intelligence systems.
State lawmaker targeted with artificial intelligence deepfake recording
A state lawmaker was impersonated using a deepfake created via Artificial Intelligence, renewing concerns about fake content in politics.
Securing artificial intelligence for America and humanity
As artificial intelligence accelerates toward unprecedented capabilities, America faces urgent choices around security and ethical safeguards.
Enterprise artificial intelligence: latest news, trends and analysis
Explore the latest developments, breakthroughs, and challenges shaping enterprise artificial intelligence—from new legislation and carbon impact to legal, security, and business implications.
Generative artificial intelligence sees widespread adoption in Swiss workplaces
Generative artificial intelligence is rapidly becoming embedded in Swiss companies, with most employees reporting positive experiences and frequent use.
Echo Chamber Attack exposes critical flaws in large language model safeguards
A new jailbreak technique known as the Echo Chamber Attack circumvents advanced large language model security, raising major Artificial Intelligence safety concerns.
Anthropic research highlights SnitchBench vulnerabilities in large language models
A new Anthropic study shows major Artificial Intelligence models can be tricked into revealing sensitive information using the SnitchBench benchmark.