A look at the targeted AI taskflows behind these findings, the bugs they uncovered, and how to run the same open-source agent on your own app.

30 techniques from conventional system engineering to supercharge AI Engineering

Technical analysis of 39+ open-source AI pentesting agents — architecture, benchmark aggregation across 8 frameworks, and tool chaining from recon to exploit.

Learn how Phantom Labs escaped the Microsoft Copilot Studio Code Interpreter sandbox, bypassed AI guardrails, and uncovered platform-wide security risks.

I gave an AI agent access to a cable modem, a serial console, Ghidra, and a custom GDB stub, then asked it to reverse the firmware and hunt for bugs. It re-discovered known vulnerabilities, found new ones, and wrote working exploits. This post covers the setup, what it managed to do on its own, and where I still had to step in. The real problem is that the same approach now works on millions of devices nobody plans to fix.

Drinor was awarded $150,000 for CVE-2026-20685 targeting Apple's Private Cloud Compute, the inference backbone of Apple Intelligence capabilities. This work is my contribution to Sentry's AI Security research initiative, run through SARC, the Sentry Applied Research Center. Our SARC focuses on evaluating systems that carry

"Over the first few weeks of Patch the Planet, our engineers found three techniques for using /goal, which hands Codex an open-ended objective and lets it work independently toward a success condition."

Unless you've been living under a rock, you're probably aware frontier large language models (LLMs) have been keenly focused on automating software development, vulnerability discovery, and most recently exploit development. The most publicized instance of this was Anthropic's marketing campaign around the release of Mythos. What better way to drum up business than by

Note: This blog post describes the simulated malicious use of unrestricted local AI large language models and is written for information…

A hands-on guide to using a local open-source LLM to write fuzzing harnesses, then finding a real memory-corruption bug with AddressSanitizer, complete with runnable code and a downloadable lab.

Hacker House provides leading cybersecurity services, expert consulting, and security training for organizations worldwide.

Trying to coerce useful work out of LLMs without the harness is like supervising a room full of drunk toddlers, each convinced they're helping, none of them checking with each other and falling over the next.

SpecterOps reverse engineered Cortex XDR with LLMs to extract YARA rules, ML models, and behavioral detections.

The CCN project is co-financed by the European Regional Development Fund and the State Budget under the European Funds for Digital Development Programme 2021-2027. Fuzzing is an automated software testing technique that involves feeding random or deliberately malformed input data to detect bugs and security vulnerabilities. For years it has …

How one Commit Broke Obfuscation: A blog post exploring the role of compilers and optimizations in the field of obfuscation and de-obfuscation.

If you would like to read the Korean version, please visit se1en.tistory.

By Tyler Holmwood on 2026-05-14

Written by Hyunseo Shin

Elastic Security Labs explores the ongoing arms race between LLM-driven reverse engineering and obfuscation.

Learn how attackers exploit tokenization, embeddings and LLM attention mechanisms to bypass LLM security filters and hijack model behavior.

Alt title: Bullying LLMs into submission to find 0days at scale

How pwn.ai turned a routine client pentest into multiple ImageMagick zero-days. By autonomously researching the product’s internals for days, it found a path from a single image upload to file leaks, security policy bypasses, and remote code execution across a wide range of default and real-world configurations.

I had $100 in unused Google Cloud credits from my Google One Ultra plan and figured I’d put them toward something interesting. Over the long weekend I ended up building an automated pipeline that scans thousands of Windows kernel drivers for exploitable vulnerabilities, specifically looking for ones that can be used in BYOVD (Bring Your Own Vulnerable Driver) attacks. On its first real run on a massive driver pack, it successfully flagged a zero-day in an ASUS driver.

After a long week, the Eye Security Research Team thought it was time for a small light-hearted distraction. So we rooted Copilot.

For decades, many bug hunters and security pros have used manual testing, static analysis, and fuzzing to find vulnerabilities. Enter the age of AI. Many are skeptical, but when I look around I see more and more evidence that it's not "going to be" a...

Collection of LLMs finetuned for bughunting

Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.

Recently I ran an experiment where I built agents on top of Opus 4.5 and GPT-5.2 and then challenged them to write exploits for a zeroday vulnerability in the QuickJS Javascript interpreter. I adde…