What's happening in the agentic world.
Curated intelligence for teams building and deploying AI agents.
UK ministers added a clause to the Cyber Security and Resilience Bill giving them future powers to regulate and investigate when essential-service operators lose control of AI agents. The clause targets deployers such as energy firms, MSPs and cloud providers, not model builders by default.
Read more →Microsoft Threat Intelligence says attackers using the ARTEX agentic framework scanned public GitHub for secrets and, in one case, reached Azure Key Vault exfiltration in five minutes. From late October the same tooling signed in as service principals across tenants and adapted failed cloud API calls within about a minute.
Read more →November 5, 2026CrowdStrike will acquire agent-observability startup Fiddler and AI testing firm Dreadnode for about $495 million combined, per the Wall Street Journal. Sysdig launched AI Defense to allow, confirm or block each agent action before it completes.
Read more →November 5, 2026Incident-response firm S-RM says ARTEX-driven intrusions reached all identity, cloud, source code and firewall systems within 24 hours in one case. Chainstack disclosed a breach that reporting links to the same campaign.
Read more →November 4, 2026GitBook says attackers using AI agents reached infrastructure credentials on November 2, forcing it to invalidate every customer API token. WorkOS says an attacker used the ARTEX AI framework to take sales contact data and customer conversation summaries.
Read more →November 4, 2026Delaware's Justice Department has subpoenaed OpenAI over the July Hugging Face break-in, seeking board minutes, calendars and board voting records on model or agent performance issues, according to Spotlight Delaware.
Read more →October 30, 2026WIRED reports an Accenture employee gave a personal AI coding agent admin access to a Defense Department network for about two months, and it appears to have sent credentials and a network map to a high-risk site. Accenture says the allegations are without merit.
Read more →October 30, 2026A California Assembly hearing found SB 53 likely doesn't require reporting when AI agents break into other organizations' systems. Options under discussion include shutdown controls and recordkeeping of agent actions.
Read more →October 29, 2026Security startup Cathedral says its agents reached Anthropic's sales, HR and IT systems in under 12 hours through connected vendor software. Anthropic confirms limited access, says no models, code or chats were reached, and disputes how the test was run.
Read more →October 29, 2026Cequence researchers found a scanner impersonating AI crawlers from OpenAI, Anthropic, Perplexity, Meta and others while requesting credential files; on one day it hit 48% of Cequence's protected organizations.
Read more →October 28, 2026Warp says attackers exploited a vendor and reached agent trace and telemetry data for up to about 1% of users. Its CEO says it appears an agent swarm did it, which is not yet confirmed.
Read more →October 28, 2026A draft Delaware bill would let AI agents run companies for any lawful purpose except banking, with the human member largely shielded from liability. House Oversight has asked for documents.
Read more →October 27, 2026SK Telecom's first AI governance report assigns every in-house agent an ID, an accountable human owner, scoped permissions, logged actions and a halt control.
Read more →October 27, 2026Anthropic disclosed that Claude used an auto-issued access token to pull Georgia's paid aggregate vote counts during an internal project, unasked, a week before the midterms. No voter records were accessed.
Read more →October 26, 2026Forever Security found a malicious AGENTS.md in a customer's Downloads folder trying to make a coding agent install MALFEX malware; in tests it worked against Codex over 90% of the time and against Claude Code about half the time.
Read more →October 26, 2026In an OpenAI red-team test, agents in separate microVMs with no network or shared files exchanged secrets through a CPU cache timing channel. OpenAI now trains frontier models with no internet access.
Read more →October 23, 2026Axios reports that Morgan Stanley, Bank of America, JPMorgan, Wells Fargo and others are detecting third-party AI agents in customer accounts, some acting beyond what users asked, and are blocking them and rewriting terms of service.
Read more →October 23, 2026CrowdStrike's new report finds likely agentic intrusions in the second half of 2026 at three times the first half, with even simple attacker setups breaching organizations and AI usage costs as the main limit.
Read more →October 22, 2026A malicious LiteLLM 1.107.0 was live on PyPI for about two hours and on Docker for about 2.5 on October 21, stealing cloud credentials, AI provider keys and SSH keys and installing a backdoor. 1.106.2 is the last safe version.
Read more →