A new decoy-counting benchmark examines whether frontier models can distinguish relevant visual information from distractions. Latent Space describes a gap between seeing and reasoning, alongside Elorian's Glimpse approach to thinking with images. The work highlights why visual recognition alone is an incomplete measure of multimodal model capability.
BleepingComputer reports that the threat group tracked as UAT-11473 exploited CVE-2026-20127 in Cisco Catalyst SD-WAN Controllers. The authentication-bypass flaw was used to deploy a custom malware framework called KeyDrift, making controller security directly relevant to network operations teams.
CISA added a GitLab remote-code-execution vulnerability and a Cisco firewall denial-of-service vulnerability to its Known Exploited Vulnerabilities catalog, according to The Hacker News. The GitLab issue, CVE-2026-97229, is described as an integer overflow or wraparound flaw with a CVSS score of 10.0. The additions signal observed exploitation rather than only theoretical exposure.
TechCrunch reports that Mistral released weights for a trillion-parameter AI model targeting cybersecurity. Mistral says the release gives organizations more control over their cyber defenses than closed models provide. That control is the company's stated rationale, rather than evidence here of independently measured security performance.