Risky Bulletin

Tom Uren and James Wilson talk about Anthropic rolling out its latest models only to have them effectively banned by the US government within days. Although the administration’s process for assessing new models is, ahem, amorphous, Anthropic is doing itself no favours by dismissing its concerns. The company needs to show some emotional intelligence and learn how to manage upwards. They also discuss Section 702 Foreign Intelligence Surveillance Act collection. The law authorising it has lapsed amidst political shenanigans, but it looks like collection can continue until next year. Plenty of time for kicking of political footballs! This episode is also available on YouTube

In this edition of Between Two Nerds Tom Uren and The Grugq talk about how NATO is set up to deter conventional conflict, and how that approach is fundamentally unsuited for ongoing, everyday cyber operations that are intended to confound adversaries. This episode is also available on YouTube.

In this Risky Business sponsored interview, Catalin Cimpanu talks with Brandon Dixon, co-founder and CTO of Ent AI, about the company’s innovative use of local LLMs to track user behavior on the endpoint, and add context to suspicious events to detect or prevent malicious activity.

In this sponsored interview Casey Ellis chats with Truffle Security’s founder and CEO Dylan Ayrey about the recent CISA secrets leak. Days after Brian Krebs ran the story, plenty of the exposed credentials were still live, including an admin-level GitHub app key with full rights over CISA’s org. Dylan walks through why deleting the repo doesn’t fix anything, why most cloud vendors won’t hard-revoke exposed keys (OpenAI and Slack will; AWS, Google and friends mostly won’t), why Hugging Face datasets now hold more secrets than GitHub itself, and what the next generation of multi-provider credential-harvesting supply chain worms is going to look like.

Iran will reconnect to the Internet, a new vulnerability lets attackers bypass authentication on AI infrastructure, hackers breach Lithuania’s state registry, security firms take down the Glassworm botnet, and CERT India releases strict patching advice.

In this sponsored interview James Wilson chats with Sondera CEO Josh Devon about why guardrails and instruction files aren’t enough to keep AI agents from going haywire. EDR, DLP and other traditional controls can’t and won’t prevent agents from going rogue. Josh explains Sondera’s “principle of least autonomy” for agents: let them do useful work, but put them in a deterministic policy harness so they can’t leak secrets, abuse tools or wander off-task.

Tom Uren and James Wilson talk about moves from several European governments to ditch Signal and set up their own encrypted messaging systems for internal government use. These efforts are motivated by concerns about phishing and sovereignty, but the solutions being adopted are imperfect and will come with their own set of problems. Signal fills a space that can’t be filled with sovereign capability. They also talk about Fast16 malware. We are only now learning about the second arm of a mid-2000s campaign to delay Iran’s nuclear weapons program that included the infamous Stuxnet worm. This episode is also available on YouTube