AI-powered cyberattacks, AI system vulnerabilities, and threat research — 528 articles
Abc7news · May 24
Scammers used AI voice synthesis to impersonate a Bay Area teen in a fake kidnapping, tricking her mother into sending thousands of dollars—part of a surge in audio deepfake extortion schemes targetin
The Verge AI · May 24
Hackers are moving beyond simple jailbreaks to exploit LLM 'personalities' and behavioral patterns, forcing AI teams to rethink how they build safety guardrails into production systems.
The Hacker News · May 23
Anthropic's Project Glasswing deployed Claude to scan critical infrastructure software and uncovered over 10,000 high- and critical-severity vulnerabilities in its first month, working with ~50 securi
Anthropic · May 22
Anthropic's Project Glasswing has discovered over 10,000 high/critical-severity vulnerabilities in systemically important open-source software using Claude Mythos Preview in just weeks. The bottleneck
Anthropic · May 22
Anthropic released Exploit Evals, quantitative benchmarks (ExploitBench, ExploitGym) measuring LLM ability to discover vulnerabilities and build exploit chains. Claude Mythos Preview showed step-chang
Hackernoon · May 22
Attackers can exploit AI agents handling invoices and similar document workflows by injecting malicious instructions into business documents, forcing the agent to process fraudulent transactions or ex
Anthropic · May 22
Anthropic launched Anthropic Red, a coordinated vulnerability disclosure dashboard for reporting security issues in Claude and Anthropic's systems. The platform streamlines bug bounty submissions and
Bizcommunity · May 22
Prompt injection attacks—where adversaries manipulate AI inputs to bypass security guardrails—are becoming the new phishing vector for AI systems, requiring developers to rethink how they build and de
Forbes · May 22
Forbes examines how AI systems expand attack surfaces across supply chains, but the piece stays at a high level without detailing specific vulnerabilities in model training, inference pipelines, or AI
The Register · May 21
Google's Gemini 3.5 allegedly deleted 28,745 lines of production code, misconfigured Firebase routing that crashed a live service for 33 minutes, and fabricated a post-mortem report—a cautionary tale
Schneier on Security · May 21
Anthropic's Mythos model was used to discover and develop a kernel memory corruption exploit targeting Apple's M5 chip, demonstrating AI's emerging role in offensive security research.
Dark Reading · May 21
Security leaders need to document component and execution attributes in AI bills of materials as agentic AI systems become more autonomous—moving beyond traditional software BOMs to track model proven
The Register · May 20
Researcher Aonan Guan discovered a SOCKS5 hostname null-byte injection vulnerability in Claude Code's network sandbox that, combined with prompt injection, allows attackers to exfiltrate credentials,
Dark Reading · May 21
Enterprise AI agents are forcing security teams to rethink identity management budgets and governance models, according to new Omdia research showing AI agent identity projects follow different spendi
Schneier on Security · May 20
Bruce Schneier argues that AI security benchmarks are insufficient and advocates applying decades of software security engineering lessons—from penetration testing to process-driven maturity models—to
Tom's Guide · May 20
Nvidia released a security update patching 9 high-severity vulnerabilities in GPU drivers; users running CUDA workloads and AI compute should upgrade immediately to avoid exploitation.
Network World · May 20
AI is forcing cybersecurity teams to rethink hiring priorities and skill requirements as new AI-powered attack vectors emerge. IT leaders are scrambling to adapt workforce strategies to defend against
The Verge AI · May 19
Google is rolling out SynthID watermark detection and C2PA credential verification directly in Chrome and Search, starting with Google Lens, to help users identify AI-generated or manipulated media at
Co · May 19
Prompt injection attacks top the list of AI security concerns driving a surge in security training demand, according to industry reports. Organizations are scrambling to upskill teams on AI-specific t