Tag: #cybersecurity

Preprint: Sparse parameter backdoors in image models can be computationally hard to detect

An arXiv preprint argues that tiny, noise-masked parameter changes can hide backdoors in pre-trained image classifiers, making tampering hard to detect.

#ai, #machinelearning, #cybersecurity, #modelsecurity

Alleged China-linked Hacker Behind HAFNIUM Extradited from Italy, Appears in Houston Court

A Chinese national accused of taking part in the HAFNIUM Microsoft Exchange campaign and stealing COVID-19 research was extradited from Italy and appeared in Houston federal court.

#cybersecurity, #china, #extradition, #hafnium

Preprint Says Federated-Learning Clients Could Remotely Trigger Rowhammer Bit Flips on Servers

An arXiv preprint claims manipulated federated-learning clients can cause Rowhammer bit flips on a server. The work is under review and unverified.

#cybersecurity, #federatedlearning, #rowhammer, #aisecurity

LayerZero links $292M Kelp DAO exploit to DPRK-linked Lazarus Group

LayerZero says a forged cross‑chain message let attackers drain 116,500 rsETH (~$292M) from Kelp DAO on April 18; private attribution points to DPRK-linked Lazarus.

#crypto, #defi, #layerzero, #cybersecurity, #lazarus

Researchers say nearly-inaudible audio can hijack voice AIs; paper claims demos on Mistral, Microsoft

A new paper says nearly-inaudible audio prompts can steer voice AI into unauthorized actions; authors report demos on Mistral and Microsoft.

#ai, #cybersecurity, #voiceai, #mistral, #microsoft

OpenAI expands Trusted Access for Cyber, unveils GPT‑5.4‑Cyber and pledges $10M in credits

OpenAI scales its Trusted Access for Cyber program, launches GPT‑5.4‑Cyber for vetted defenders and commits $10M in API credits to security teams.

#openai, #cybersecurity, #ai, #gpt5

ShinyHunters Leak Linked to Rockstar Mostly Reveals Cloud Analytics, Not GTA 6 Assets

ShinyHunters exposed Rockstar-related cloud analytics via an Anodot–Snowflake integration on April 13, not GTA 6 code or player data.

$TTWO, #cybersecurity, #gaming, #cloud, #data-breach

Anthropic’s New AI Finds Thousands of Vulnerabilities, Launches Project Glasswing

Anthropic says its Claude Mythos model uncovered thousands of previously unknown software flaws; Project Glasswing gives vetted partners guarded access for defensive work.

#ai, #cybersecurity, #open-source, #anthropic