GitLost: How Prompt Injection Leaks Private GitHub Data
534
GitHub's AI agents can be manipulated through public issues to leak private repository data, highlighting a major security flaw in agentic workflows.
Vulnerability Research
Automated and manual discovery of software vulnerabilities, exploit development, and security testing against real-world targets.
Reading List
Open-Weight GLM 5.2 Beats Claude in Semgrep Cyber Benchmarks
1098
Open-weight model GLM 5.2 surpassed frontier models in IDOR detection benchmarks, signaling a shift toward cost-effective and private AI for security tasks.
Exploitarium: A Consolidated Archive of Public Vulnerability Research and PoCs
942
Exploitarium is a consolidated open-disclosure archive of vulnerability research and exploit PoCs designed for educational purposes and community reporting.
Mythos: AI's New Frontier in Automated Exploitation and Defense
171
Claude Mythos accelerates the threat of automated cyberattacks, making the adoption of Zero Trust and AI-assisted defense an urgent necessity rather than an option.
Mythos vs. The World: Benchmarking AI in Security Bug Hunting
319
A benchmark of public AI models reveals that Anthropic's Mythos is uniquely skilled at finding elusive security bugs, though cheap Chinese models are rapidly closing the gap.
The 'Fix This Code' Ban: Why Feds Blocked Anthropic's Latest AI
606
A US government ban on Anthropic's advanced AI models is being criticized by experts as a misguided reaction to standard defensive coding prompts.
SkillSpector: NVIDIA's Security Scanner for AI Agent Skills
SkillSpector is an automated security tool that scans AI agent skills for vulnerabilities and malicious intent using static and semantic analysis.
Anthropic's Reference Harness for Autonomous Vulnerability Remediation
532
An open-source reference implementation for building autonomous, LLM-powered vulnerability detection and remediation pipelines.
LLM Hacking Trial: GPT-5.5 Dominates in $1,500 Firebase Exploit Test
400
An evaluation of various LLMs found that GPT-5.5 is highly effective at exploiting Broken Access Control vulnerabilities, though safety filters and high costs remain significant barriers for other models.
VSCode Webview Flaw Enables GitHub Token Theft via github.dev
658
A flaw in VSCode's keyboard event handling allows malicious repositories to steal GitHub tokens on github.dev by automatically installing rogue extensions.
Anthropic Expands Project Glasswing to Secure Global Infrastructure with AI
180
Anthropic is scaling its AI-driven cybersecurity initiative, Project Glasswing, to help critical infrastructure providers and open-source maintainers detect and patch vulnerabilities before AI-powered attacks become widespread.
Project Glasswing: AI Finds 10,000 Vulnerabilities in One Month
549
Project Glasswing demonstrates that AI can find software vulnerabilities at an unprecedented scale, shifting the security focus from discovery to the urgent need for faster patching.
Scaling AI Vulnerability Research: Lessons from Project Glasswing
360
Cloudflare’s research with Mythos Preview demonstrates that while AI can autonomously chain exploits, effective defense requires specialized multi-agent harnesses and a focus on architectural security.
Pause Software Installations Amid New Linux Vulnerabilities
854
Due to new Linux kernel vulnerabilities, users should avoid installing new software for a week to prevent supply chain attacks.
Critical RCE Vulnerability Discovered in GitHub's Internal Git Infrastructure
446
Wiz Research used AI-augmented tools to find a critical RCE vulnerability in GitHub's internal protocol that could compromise millions of repositories via a simple git push.
AI-Driven Audit Secures OpenEMR Against 38 New Vulnerabilities
177
AISLE used autonomous AI analysis to discover and help patch 38 vulnerabilities in OpenEMR, establishing a new standard for proactive healthcare software security.
GPT-5.5: A Step Change in AI-Powered Hacking
GPT-5.5 delivers a revolutionary increase in vulnerability detection and hacking efficiency, outperforming previous models and setting a new bar for AI in cybersecurity.
Intelligence, Not Compute, Will Win the AI Cybersecurity Race
237
AI cybersecurity is a contest of model intelligence and reasoning, not a brute-force competition of computational resources.
AI Achieves Root Access on Samsung Smart TV
255
OpenAI's Codex successfully discovered and exploited a kernel memory vulnerability to gain root access on a Samsung Smart TV.
The Ransomware Gap: Activity Outpaces Security Spending 3-to-1
Ransomware activity is currently outpacing global security spending growth by a factor of three to one.
The Essential Plugin Attack: 30+ WordPress Plugins Weaponized via Supply Chain
1182
A malicious actor weaponized a portfolio of 30+ acquired WordPress plugins to conduct a massive, blockchain-coordinated supply chain attack.
The Benchmark Illusion: How UC Berkeley Broke the World's Top AI Leaderboards
523
Current AI agent benchmarks are easily gamed through infrastructure exploits, necessitating a new standard of adversarial robustness and environment isolation to accurately measure model capabilities.
The System is the Moat: Why Small Models Rival Frontier AI in Cybersecurity
1268
AI cybersecurity is a 'jagged frontier' where small models often match frontier performance, proving that the orchestration system is the true competitive moat.
Anthropic Restricts Claude Mythos to Prevent AI-Driven Security Crisis
Anthropic is restricting its powerful new Claude Mythos model to a select group of security partners to prevent a potential wave of AI-driven cyberattacks while patching critical software vulnerabilities.
Project Glasswing: Securing Global Infrastructure with Frontier AI
1517
Project Glasswing is a collaborative effort to use Anthropic's highly capable Claude Mythos model for defensive cybersecurity to protect critical global infrastructure from AI-augmented threats.
Claude Code Unearths 23-Year-Old Linux Kernel Vulnerability
424
Anthropic researcher Nicholas Carlini used Claude Code to uncover a 23-year-old Linux kernel vulnerability, signaling a new era of AI-driven security research.
OpenClaw Fixes High-Severity Privilege Escalation Vulnerability
502
OpenClaw version 2026.3.28 fixes a critical authorization flaw that allowed users to escalate their privileges to admin via the device pairing process.
The Death of the Security Embargo: Navigating the Kernel's Bug Surge
313
A massive influx of valid security reports is ending the era of secret embargoes and forcing a shift toward continuous software maintenance.
Snowflake Patches Critical Sandbox Escape and Malware Execution Flaw in Cortex AI
266
Snowflake Cortex Code CLI was vulnerable to a sandbox escape and human-in-the-loop bypass that allowed unauthorized malware execution via indirect prompt injection.
Vetting the Blast Radius: The AI Skills Security Index
A security database that evaluates and ranks the instructional risks and permission levels of AI agent skills to prevent exploitation.
Autonomous AI Agent Breaches McKinsey’s Lilli Platform
499
An autonomous AI agent hacked McKinsey’s internal AI platform in two hours, exposing millions of confidential records and highlighting the urgent need to secure the prompt layer.
Claude AI Accelerates Firefox Security Research
628
Claude Opus 4.6's discovery of 22 Firefox vulnerabilities highlights a powerful, yet potentially temporary, AI-driven advantage for software defenders.
Exploits at Scale: When Token Throughput Becomes the Bottleneck
265
Exploit development is becoming a token-limited, scalable process with LLMs, so we must prepare and demand real-target, high-budget evaluations.
Notion AI Pre-Approval Edits Enable Prompt-Injection Data Exfiltration
206
Notion AI saves edits before consent, enabling prompt-injected external image loads that exfiltrate user data regardless of user approval.
Supply‑Chain XSS in Mintlify Let Attackers Run JS on Discord, X, and More
1167
An exposed Mintlify static endpoint let malicious SVGs run on customer primary domains, creating a widespread supply-chain XSS affecting Discord, X, and many others.
OpenAI Launches GPT-5.2-Codex for Advanced Agentic Coding and Cyber Defense
589
OpenAI’s GPT-5.2-Codex pushes agentic coding and defensive cyber forward while rolling out with stricter safeguards and gated access.
React RSC: Update Now to Patch DoS and Source Leak (CVE-2025-55184, -55183)
346
Update your RSC stack now: fixed react-server-dom versions patch a DoS and source code leak that affect many frameworks, though no new RCE is possible.
Critical Next.js RCE via React Server Components—Update Now
628
Critical RCE in React Server Components affects Next.js App Router; upgrade to the listed patched versions now.
First AI-Agent Orchestrated Cyber Espionage Disrupted; Defense Must Adapt
376
AI agents have enabled near-autonomous, state-linked cyber espionage at scale, forcing a rapid shift toward AI-powered cyber defense and stronger safeguards.