One question to an AI agent may set off hundreds of data requests, says @nightdragon's Morgan Kyauk, and each one needs its own permission check. To me, that means keeping the rules on the data, so the check travels with it. nightdragon.com #DataCentricSecurity

Only 14% in @VentureBeat's research said a named owner could cut off an AI agent's access within minutes. Nebius's Nir Chervoni says each agent needs its own identity and task-scoped credentials. I'd check that entitlement at the data, with ABAC and TDF. venturebeat.com/video/why-the-…

An AI agent's instructions are only suggestions, my colleague Don Yeske says on @illumio's podcast, alongside John Kindervag, who created Zero Trust. Real limits are ones the agent can't undo. @virtruprivacy #ZeroTrust illumio.com/podcast/zero-t…

Here's the teaser I used to introduce colleagues at @virtruprivacy to Virtru Decks. I think the best AI tools marketing builds are ones anyone can use. This one reads our brand and messaging context, so a seller can build a deck faster, on brand and on message. #AgenticMarketing

Once coding and review agents trade comments, a pull request becomes a log only agents read, argues @GoEntire's Thomas Dohmke. At Anthropic, AI writes about 80% of merged code, and people set the intent and own final approval. entire.com/blog/the-git-f…

Almost every organization in Delinea's 2026 report has a policy on what data AI agents may access. Only 51% check access against it in real time, Matt Howard of @virtruprivacy notes. What made real-time checks practical for your team? #AIGovernance virtru.com/blog/data-cent…

Giving each AI agent its own identity is overdue. @SailPoint found 2% of organizations use identity tools built for agents. I'd add that an agent can drift with every permission valid, so the data needs to protect itself. @virtruprivacy sailpoint.com/horizons

A classified conversation is data too, Will Peppo of @virtruprivacy says, so access to it should be checked continuously. With @pexip, it's checked when a meeting is scheduled and again as each person joins. I'm proud we partner on that. #ZeroTrust virtru.com/newsroom/virtr…

Microsoft offered buyouts to about 60,000 employees. Its Copilot and CoreAI teams are eligible, but its security teams aren't, per The Verge's sources. My read: as agents take on more of the work, deciding what each can touch becomes a bigger job. theverge.com/tech/1018666/m…

This CyberSecurity Awareness Month, add AI agents to the insider-risk list. One can be fully auth'd and still leak sensitive data while chasing its goal, Wayne Chung of @virtruprivacy tells @vmblog. How does your team review what agents share? #InsiderRisk buff.ly/Y5YtMAI

Keeping AI agents away from real data makes them safe, but not much use. Nick Michael of @virtruprivacy makes the case for enforceable limits on the data itself, so agents can do real work inside them. Which files did your team trust an agent with first? virtru.com/blog/data-cent…

My colleague Eugene Yakhnenko says some contributors to his open source editor talk to him through their AI agents. I think agents will soon send more PRs than people, if they don't already. Which CI check do you trust most with an agent's PR? github.com/eugenioenko/ttt

An agent's permissions can change mid-job. OpenAI's Siddarth Jain says controls should keep rechecking them, per @techtargetnews. To me, agents can drift without breaking a rule, so the data has to protect itself. @virtruprivacy #AIGovernance techtarget.com/data-technolog…

Customers stick with security that doesn't slow their work down. @virtruprivacy is an @Inc Power Partner for a fifth straight year, and John Ackerly says businesses should never have to choose between protecting sensitive data and putting it to work. virtru.com/newsroom/virtr…

A reporting agent has no reason to open payroll. On an analyst's login, it still passes both the identity check and the data check. SACR argues identity and data need to be judged together, while the agent is working. #IdentitySecurity softwareanalyst.substack.com/p/the-rise-of-…

An AI agent can pull from two well-guarded systems and post the result somewhere neither can see. Jeremy Bayer argues in RealClearDefense that the question now is how to protect data that has to cross many walls. How are you tracking agent output?

buff.ly/Z8S8yor

I'm a hands-on learner, so I really appreciate what my colleague Devin Collins built. He made it so his dad, who learns the same way, could watch GPT-2 work through a question step by step.

Who in your life keeps asking how AI works?

buff.ly/VvCwAAv

A newly cleared person can wait weeks while each tool's admin adds them by hand. JP Ayyappan of Virtru says one attribute change opens all of a program's material in Rocket.Chat, per Virtru's blog. @virtruprivacy #ABAC buff.ly/SPn0Lx6

On Oct 26, @DMV_Rising (co-sponsored by @virtruprivacy) is Day 1 of DC's new Cyber Innovation Conference. Startups pitch investors at the @DataTribe Challenge on Day 2. It's a smart pairing for the region's cyber scene. dmvrising.com

Keynotes come from the founders of @xbow and @CapeCellular. Thanks to sponsors @anacondainc, @ExpelSecurity, @sonatype, @method_security and @threater_inc, media partner @CyberEdBoard, and community partners @the_cyberguild and @DCALive.