Lakera bulletin - This week in AI - Nov 10, 2025

_Val_ — Mon, 10 Nov 2025 16:14:02 GMT

From adaptive AI-powered malware to new tools for safer model outputs, this week brought big movement across AI security and open innovation. Companies are also waking up to the reality of AI risk management, and realizing they’re not quite ready yet.

Let’s jump right in.

Malware Now Uses AI to Mutate in Real Time

Google’s Threat Intelligence Group has uncovered malware that embeds AI models directly into its payloads, allowing it to adapt its code, behavior, and data collection dynamically during execution. This marks a significant shift in adversarial use of AI, with implications for both detection and response strategies.
🔗 Read the full report

OpenGuardrails Launches to Detect Unsafe AI Outputs

A new open-source project, OpenGuardrails, offers contextual guardrails to detect and mitigate unsafe model outputs: from prompt injections to privacy leaks. Released under Apache 2.0, it’s designed to make AI systems more trustworthy and compliant by default.
🔗 See the announcement

Pinterest CEO Highlights Cost Savings from Open-Source AI

Pinterest CEO Bill Ready revealed that open-source AI models are delivering “orders of magnitude” cost reductions compared to proprietary ones, especially in visual search. The move highlights how open systems are reshaping enterprise AI strategies by cutting costs while maintaining performance.
🔗 Read the story

Companies Admit They’re Under-Prepared for AI Security Risks

A new industry report shows most organizations adopting AI are struggling to keep pace with security and governance demands. Lakera’s own 2025 GenAI Security Readiness Report echoes the trend: only 19% of enterprises describe their GenAI security posture as “highly confident.” As adoption surges, it’s clear that AI security maturity is lagging behind deployment.
🔗 Read the analysis

Google Recaps Major AI Updates in October

Google’s latest AI roundup includes advances in cancer detection, a new quantum algorithm, and enterprise AI integrations. The update shows how frontier AI research continues to intersect with applied domains like healthcare and scientific discovery.
🔗 See Google’s blog

Organizations Ramp Up AI Oversight, but Gaps Remain

A new report finds enterprises are increasing budgets for AI oversight and governance, yet few have the maturity to securely manage these systems at scale. The research suggests that AI risk management is gaining traction, but tooling and training still lag behind deployment.
🔗 Read the briefing

From self-modifying malware to open-source safety layers, the week showed both the creative potential and the security stakes of AI’s rapid evolution. The message is clear: innovation is accelerating, but governance needs to keep pace.

Re: Lakera bulletin - This week in AI - Nov 10, 2025

the_rock — Mon, 10 Nov 2025 16:25:25 GMT

Love these posts about Lakera!

topic Re: Lakera bulletin - This week in AI - Nov 10, 2025 in AI Agents Security

Lakera bulletin - This week in AI - Nov 10, 2025

Malware Now Uses AI to Mutate in Real Time

OpenGuardrails Launches to Detect Unsafe AI Outputs

Pinterest CEO Highlights Cost Savings from Open-Source AI

Companies Admit They’re Under-Prepared for AI Security Risks

Google Recaps Major AI Updates in October

Organizations Ramp Up AI Oversight, but Gaps Remain

Re: Lakera bulletin - This week in AI - Nov 10, 2025

Re: Lakera bulletin - This week in AI - Nov 10, 2025