October 26 to November 1, 2026
period in progress
News
php2shell groups the critical PHP flaws CVE-2026-109618 and CVE-2026-109619.
Claude Mythos 5.1 solves the task in 3h 28m and GPT-5.6 Sol in 6h 10m.
GLM 5.3 gets code execution locally but fails against the remote target after more than 20 hours.
Kimi K3.1 Pro and DeepSeek V4.1 Flash never reach the leak of the exploit's first step.
News
The ReproBAIT evaluation, built by the AISI with SecureBio, provides the training data and a single offline GPU.
On ProteinMPNN, Claude Opus 5 reaches 96.5% fidelity with the full paper, but RFdiffusion3 resists the tests.
According to the AISI, withholding the code of a tool therefore does not on its own guarantee safe release.
News
An attacker chooses fields such as a User-Agent or a DNS query that the agent reads without a separate channel.
The false authority makes 68% of malicious logs pass as benign against a naive prompt.
Context manipulation compromises 96% of summaries without a defense.
The best defense brings the average success from 26.6% down to 11.8% in tests on the gpt-4o-mini model alone.