Insecure Agents Podcast

131 posts

Insecure Agents Podcast banner
Insecure Agents Podcast

Insecure Agents Podcast

@insecureagents

Insecure Agents is a podcast and brand that curates the most important conversations and trends in AI security. Hosted by @vtahowe, co-founded by @kylebhiro.

San Francisco, CA Katılım Haziran 2025
92 Takip Edilen766 Takipçiler
Insecure Agents Podcast
Insecure Agents Podcast@insecureagents·
Guillermo reports "we believe the attacking group to be highly sophisticated and, I strongly suspect, significantly accelerated by AI. They moved with surprising velocity and in-depth understanding of Vercel" Alex Stamos warns us that defensive agents with autonomy and capability need to be part of our security response. Humans and human in the loop are too slow. "If you have to wake up a person at 3 a.m. and they have to open their laptop and log in and go through and like figure out, God, what's going on? Okay, yes, I approve my AI agent doing something radical like turning off an account or shutting down a container. That's too late, right? Because if that malware is being driven by AI, it's only gonna need a couple of minutes to do that kind of internal exploration to pick out an exploit to make the next jump. And you're already behind the ball, you're chasing them. " - @alexstamos
Guillermo Rauch@rauchg

Here's my update to the broader community about the ongoing incident investigation. I want to give you the rundown of the situation directly. A Vercel employee got compromised via the breach of an AI platform customer called Context.ai that he was using. The details are being fully investigated. Through a series of maneuvers that escalated from our colleague’s compromised Vercel Google Workspace account, the attacker got further access to Vercel environments. Vercel stores all customer environment variables fully encrypted at rest. We have numerous defense-in-depth mechanisms to protect core systems and customer data. We do have a capability however to designate environment variables as “non-sensitive”. Unfortunately, the attacker got further access through their enumeration. We believe the attacking group to be highly sophisticated and, I strongly suspect, significantly accelerated by AI. They moved with surprising velocity and in-depth understanding of Vercel. At the moment, we believe the number of customers with security impact to be quite limited. We’ve reached out with utmost priority to the ones we have concerns about. All of our focus right now is on investigation, communication to customers, enhancement of security measures, and sanitization of our environments. We’ve deployed extensive protection measures and monitoring. We’ve analyzed our supply chain, ensuring Next.js, Turbopack, and our many open source projects remain safe for our community. The recommendation for all Vercel customers is to follow the Security Bulletin closely (vercel.com/kb/bulletin/ve…). My advice to everyone is to follow the best practices of security response: secret rotation, monitoring access to your Vercel environments and linked services, and ensuring the proper use of the sensitive env variables feature. In response to this, and to aid in the improvement of all of our customers’ security postures, we’ve already rolled out new capabilities in the dashboard, including an overview page of environment variables, and a better user interface for sensitive env var creation and management. As always, I’m totally open to your feedback. We’re working with elite cybersecurity firms, industry peers, and law enforcement. We’ve reached out to Context to assist in understanding the full scale of the incident, in an effort to protect other organizations and the broader internet. I also want to thank the Google Mandiant team for their active engagement and assistance. It’s my mission to turn this attack into the most formidable security response imaginable. It’s always been a top priority for me. Vercel employs some of the most dedicated security researchers and security-minded engineers in the world. I commit to keeping you updated and rolling out extensive improvements and defenses so you, our customers and community, can have the peace of mind that Vercel always has your back.

English
1
2
21
7.5K
Insecure Agents Podcast
Insecure Agents Podcast@insecureagents·
Agent native startups are uniquely positioned to deliver highly customizable software. "We let users install additional CLI tools into their sandbox environment, for example if you want Sazabi to be able to introspect your AWS configuration you can add the AWS CLI" @shcallaway
English
1
2
11
2.9K
Insecure Agents Podcast
Insecure Agents Podcast@insecureagents·
Anthony’s thread on this is worth a read
Anthony Shew@anthonysheww

After some more tinkering, we've arrived. 95% faster in our largest repo. Booyah achieved. In eight days, I did something I estimate would have taken me at least two months if I had done it totally by hand. I acknowledge that I'm not the system developer that I want to be yet. Currently, I'm still silently looking up to and watching performance PRs from greats like @boshen_c on oxc or @alistaiir, @jarredsumner, @dylanconway111 on Bun. But I'll be damned if this didn't feel good. With or without AI, I wouldn't have been capable of much of this even six months ago. And it was clear I was still leading the agent, not the other way around.

English
1
0
1
253
Insecure Agents Podcast
Insecure Agents Podcast@insecureagents·
Anthony made Turborepo 95% faster. Sandboxes played a key role in how. "That sandbox proved to be very useful from getting things from being like oh ok this is kind of fast all the way out to wow, don't blink you're going to miss it" @anthonysheww
English
1
1
27
2.7K
Insecure Agents Podcast
Insecure Agents Podcast@insecureagents·
"Agents can interpret large amounts of unstructured data in a way that humans could never do efficiently before" @shcallaway This is the unlock powering the idea of context graphs, which is only possible if agents have the autonomy, capability, and security to discover data.
English
0
1
6
4K
Insecure Agents Podcast
Insecure Agents Podcast@insecureagents·
Shared memory is both an engineering and security challenge How do you persist memory across ephemeral agent runs in sandboxes? How do you manage access to read and write from a shared memory store? "What we do is we literally just git push to that branch at the end of every sandbox execution. And that ensures that if there were any changes to the file system, they are persisted to the remote git server. And then the next time an agent runs, it pulls down whatever the latest state is for its sandbox. And this is how we share memory across the agent runs." @shcallaway
English
3
4
52
29K
Insecure Agents Podcast
Insecure Agents Podcast@insecureagents·
Yes, we should be building for agents, but developer experience still matters. "There's this weird overlap between developer experience and agent experience that is like, can I get my coding agents to go explore these different options and then come back to me with enough information and findings that gives me confidence as an architect, as a builder to make the right decision" @dexhorthy
English
2
4
18
8.8K
Insecure Agents Podcast
Insecure Agents Podcast@insecureagents·
At Daytona Compute, we explored a key nuance: Sandbox isolation techniques vary and the implementation (containers vs microVMs, etc.) directly affects your security posture. “It's a very common misconception that Docker containers provide perfect isolation.” @renebrandel
English
1
6
12
3.1K
Insecure Agents Podcast
Insecure Agents Podcast@insecureagents·
Only a handful of companies used to have to worry about a zero day. Now it's everyone. Alex Stamos tells us how model advances have made it faster and cheaper to find vulnerabilities- some even the best security reachers couldn't find. Project Glasswing is indeed urgent.
Anthropic@AnthropicAI

Introducing Project Glasswing: an urgent initiative to help secure the world’s most critical software. It’s powered by our newest frontier model, Claude Mythos Preview, which can find software vulnerabilities better than all but the most skilled humans. anthropic.com/glasswing

English
0
0
5
317
Insecure Agents Podcast
Insecure Agents Podcast@insecureagents·
Our conversations from RSAC are LIVE The AI-Driven Kill Chain and the Coming Bug Apocalypse, Alex Stamos OpenAI Acquires Promptfoo, Ian Webster From Point-in-Time Audits to Continuous Testing w Kyle Bhiro How Security Changes When Most Product Users are Agents w Mark Dorsi
English
1
0
2
670