Latent Space reports on a new decoy-counting benchmark that exposes a gap between visual recognition and reasoning in frontier models. Elorian's Glimpse explores thinking with images as a way to bridge that gap, making the work relevant to evaluations of visually grounded AI.
CISA added VMware ESX vulnerability CVE-2026-41709 to its Known Exploited Vulnerabilities catalog, according to The Hacker News. CrowdStrike linked exploitation to the China-nexus actor Warp Panda in attacks on the U.S. legal sector, with the flaw used to reduce audit-trail visibility.
Cisco released fixes for CVE-2026-76418 in the Remote Access SSL VPN service of ASA and FTD software. The flaw is rated 8.6 and is being exploited in the wild; reporting describes unauthenticated remote attacks that can crash vulnerable firewall devices.
Google released Gemini 4 Carbon, claiming significant gains in coding, agentic tasks, and cybersecurity, The Verge reports. Google's announcement highlights the model's leading position on the Artificial Analysis Intelligence Index v5.0, adding another benchmark claim for teams comparing frontier models.
Anthropic launched tools intended to show companies what their AI spending is buying, TechCrunch reports. The release focuses on visibility into how customers spend money on its models, a practical concern for teams assessing AI usage and costs.