ReversingLabs found LLM API references in malicious files rose from roughly 150 in 2024 to 1,680 in 2026, with no evidence of an autonomous AI attack surge. AI appears in individual attack stages: code rewriting, UI navigation and vulnerability discovery. PHP2Shell exploit files appeared within three days of disclosure, with 5–12% of antivirus engines detecting them.
Vulnerability Research
8 posts
The AI Cyberattack Surge Hasn't Arrived. But the Warning Signs Are Growing. Kernel eviction notice: What wesp.sys reveals about the Windows Endpoint Security Platform Microsoft's WESP framework moves EDR out of the Windows kernel via wesp.sys, but preview binaries show user-mode clients cannot block process creation yet. WESP uses a file system mini-filter (wesp.sys) at altitude 329500 communicating with user-mode clients via espclient.dll over a FilterCommunicationPort named \EspFilterPort. The client DLL imports FilterGetMessage and FilterSendMessage but not FilterReplyMessage, indicating one-way notification without blocking capability. EDR vendors must wait for Microsoft to add reply functionality before migrating blocking logic to user mode.
How we use /goal to find bugs in Patch the Planet This article is a methodology writeup from Trail of Bits describing how they leverage OpenAI Codex's '/goal' autonomous agent feature within the 'Patch the Planet' initiative to conduct large-scale vulnerability research and variant analysis against major open-source projects (Rust, curl, zlib, Keycloak, kubernetes-client). It details prompt-engineering lessons learned (letting the model draft its own goal, defining outcomes precisely without prescribing method, and assigning one outcome per agent) and describes an automated Rust P-critical variant-analysis pipeline with a two-pass validation gate before human review. No specific exploit details, IOCs, or CVE numbers for the discovered bugs are disclosed in this post.
Field reports from Patch the Planet Trail of Bits and OpenAI's Patch the Planet initiative used GPT-5.5-Cyber to autonomously build a bespoke fuzzing harness for zlib in a single day, discovering multiple vulnerabilities now undergoing coordinated disclosure. The model independently chose dynamic fuzzing over static review, wrote C/C++ harnesses across a dozen entrypoints, used ASan/UBSan builds and compile-time variants, and demonstrated disciplined reporting by filtering out unreachable crashes. This represents a significant shift in the threat model: the expertise barrier for bespoke fuzzing has collapsed, making large-scale vulnerability discovery accessible to both skilled researchers and low-skill attackers.
Introducing Patch the Planet Trail of Bits introduced the 'Patch the Planet' initiative, leveraging frontier AI models to identify and remediate vulnerabilities across critical open-source projects. The effort highlights a paradigm shift where AI accelerates bug discovery, making triage, patching, and disclosure the primary challenges for maintainers.
DICOM, Pydicom, GDCM, and Orthanc: A technical tour of what really happens in the heap Security research highlights a heap overflow vulnerability within DICOM parsing, specifically targeting Orthanc servers during image uploads. By exploiting the complex DICOM file format, attackers can trigger an out-of-bounds write, posing a significant risk to hospital PACS systems that automatically ingest and decode these files.
Extending Ruzzy with LibAFL Trail of Bits detailed the technical process of integrating the LibAFL fuzzing engine into Ruzzy, their coverage-guided fuzzer for Ruby. The integration required resolving ELF linker constraints with .preinit_array sections and adjusting shared object loading to satisfy LibAFL's strict coverage map initialization requirements.
On the Effectiveness of Mutational Grammar Fuzzing The article details the limitations of mutational coverage-guided grammar fuzzing, specifically its tendency to produce similar samples and struggle with complex function chaining. To mitigate this, the author introduces a methodology using the Jackalope fuzzer that periodically restarts workers to combine generative and mutational fuzzing, significantly improving the discovery rate of unique crashes in targets like libxslt.