METR demonstrated a stored cross-site scripting weakness in an isolated staging copy of the Inspect evaluation viewer. A model-produced payload could alter the reviewer’s displayed transcript and intercept a download, while leaving the underlying database log unchanged. The report does not describe observed exploitation during evaluations. It documents the initial June fix and later hardening that supports plain-text viewing of untrusted content.
Transluce’s September 30 follow-up analyzes public web-archive and URL-scanning records of apparent agent activity. It identifies failed SQL-injection attempts against US education data and unsuccessful probes against Library and Archives Canada, alongside aggressive retrieval of public information. Some traffic connects to benchmark tasks or OpenAI markers, but the researchers do not attribute every incident to one developer and do not confidently attribute the Canadian probes.
Play video
Charlie Holtz describes Conductor’s approach to coordinating coding agents through shared cloud workspaces and organizational context. His examples include CI-enforced human review for migrations, carefully maintained agent instructions, and a database of working information queried by agents. The presentation combines workflow advice with product demonstrations; it does not measure a general productivity advantage.
Play video
Cody Menefee proposes an AI-assisted pasture-rotation system combining grazing knowledge, field observations and GPS-collar controls. Cameras, drones or satellite imagery could supply changing pasture conditions, while an LLM recommends moves for a farmer to accept or reject. The talk identifies unresolved data and hardware-integration needs; improved farm capacity is a project goal rather than a demonstrated outcome.
We are proud to announce that Forrester has named Google a Leader in The Forrester Wave™: External Threat Intelligence Service Providers, Q3 2026.
Irregular shows that a maintenance agent with access to training tools, model weights and deployment could update a shared model. Controlled tests recovered synthetic secrets and removed a learned refusal policy; these demonstrate a mechanism, not its prevalence in deployments.
CVE-2026-65105 combines NemoClaw's non-loopback Ollama binding with DNS rebinding and an unauthenticated model API. A malicious page can reach the local runtime and alter a model's chat template so hidden instructions persist across later agent conversations; the disclosure reported platform-specific remediation limitations.
Learn more about how Google Cloud empowers you with the tools, governance, and infrastructure you need to securely deploy workloads and maintain long-term trust.
Learn how better questions, trusted AI, and human judgment help security leaders make confident decisions and build resilient systems.
Join Microsoft Security at Black Hat USA 2026 for supply chain research, hands-on security experiences, expert conversations, and our reception.
Francis deSouza explains the crucial role that deep context plays in creating an AI advantage for defenders.
Microsoft’s latest Secure Future Initiative report outlines progress on secure foundations, AI-powered defense, and future-ready cybersecurity.
Cloudflare report on testing security-focused frontier models against real infrastructure code. Relevant to evaluating AI-assisted vulnerability discovery and production security workflows.
Play video
NDC Security 2026 talk on securing agentic systems through reference architecture, common risk analysis, and control prioritization for autonomous agents in the software development lifecycle.
This post dives deep into how Claude wrote an exploit for one of the vulnerabilities it found in Firefox.
Play video
NDC AI 2025 talk introducing AI security for developers, including model lifecycle, training data, secure integration, data leakage, prompt injection, adversarial inputs, and model bias.
Google DeepMind’s December 2025 FACTS suite evaluates factual answers under four different information conditions: model knowledge alone, web search, images and supplied documents. Its search track standardizes the retrieval tool across models, while public examples accompany a private held-out evaluation set managed by Kaggle. The combined score averages results across tracks and sets, which can conceal sharply different failure patterns.
The SIMA 2 technical report describes three ways to score agent tasks in virtual worlds: environment-state checks, programmatic checks over screens and actions, and human review of recorded trajectories. Its evaluation refinements are transferable: require a success signal to persist, limit unnecessary actions after completion, and require every step of a sequential task to succeed.
Play video
This AI Explained video reviews a major AI development through the lens of AI safety and model behavior. It is useful context for AI engineering, evaluation, governance, and operational risk.
Play video
This AI Explained video reviews a major AI development through the lens of governance and responsible deployment. It is useful context for AI engineering, evaluation, governance, and operational risk.
Play video
This AI Explained video reviews a major AI development through the lens of benchmarks and evaluation evidence. It is useful context for AI engineering, evaluation, governance, and operational risk.
Play video
This AI Explained video reviews a major AI development through the lens of benchmarks and evaluation evidence. It is useful context for AI engineering, evaluation, governance, and operational risk.
Play video
This AI Explained video reviews a major AI development through the lens of benchmarks and evaluation evidence. It is useful context for AI engineering, evaluation, governance, and operational risk.
Play video
This AI Explained video reviews a major AI development through the lens of benchmarks and evaluation evidence. It is useful context for AI engineering, evaluation, governance, and operational risk.