CEO @trailofbits, organizer @EmpireHacking
Miami, FL
- Tweets 11,130
- Following 1,301
- Followers 26,814
- Likes 7,667
I started blogging again! Short posts while I get my writing muscles strong again.
Here's one about a scholarship program for hackers that I started at my high school (@mineolahs): illusivedan.com/writing/hack…
6
3
55
6,763
Dan Guido retweeted
This divide between AI safety and cybersecurity is fascinating — and important. In fact, I dug into it in depth for my newsletter Ground Level AI back in early September, interviewing more than a dozen people across both communities in the wake of the OpenAI/Hugging Face incident, discussing their very different approaches to AI risk. Including @ZackKorman, who is quoted in this @TIME piece.
groundlevel-ai.com/p/ai-safe…
I’ve been trying to make sense of this rift between the AI safety vs security camps.
time.com/article/2026/10/20/…
2
3
20
1,961
Dan Guido retweeted
Giving AI access to your money? Start by limiting the blast radius.
@anishrnaik, Principal Security Engineer at @trailofbits, on giving AI agents access to wallets, code and infrastructure.
00:00 Meet Anish from Trail of Bits
02:46 Patch the Planet
05:06 Package managers and hidden dependencies
09:34 Dependency alerts and package selection
12:27 Limiting an AI agent’s blast radius
15:47 Can AI fix bugs without creating new ones?
21:54 Hardware wallets and the wider security stack
26:43 AI attackers and defenders
28:53 OpenAI reports and what comes next
1
3
684
Dan Guido retweeted
So many people discovering that the whole Internet is dry tinder and the only thing keeping it from burning is the fact that teenagers haven’t found the matches yet.
10
36
238
10,846
Dan Guido retweeted
Our team worked with OpenCode to have their users who are also OpenAI API customers back in business and any credits refunded which would have been drawn in the attack.
Also, you can request Daybreak Blue access to use our cyber permissive setup for defense. Do not wait until you have to deal with an incident. We’re here to help.
463
293
8,204
1,276,029
I'm back at @HarvardHBS for the final unit of OPM, and there's one thing different this year: I have agents to help manage @trailofbits. I hardly need to check my email or Slack or at all between Hermes and my Dot.
Personal news: I'm taking September off from @trailofbits to attend @HarvardHBS full-time as part of their OPM program. I'll be taking classes for 8 hrs, 6 days per week so you'll be seeing fewer security tweets from me! 🧵
9
26
3,383
Dan Guido retweeted
Two reasons why I love Blindsight by Peter Watts:
1. He wanted readers to experience what fighting against superintelligence would *feel* like. Most of the novel feels like a haze, where the reader understands maybe <30% of what's happening. It's an amazing experience because he encourages readers to try their best at following along while simultaneously destroying their ability to keep up. It's like playing chess against AlphaZero: you try your best and yet you fall further and further behind without understanding why.
2. As a hard core biologist by training, he wanted to explore: what is the evolutionary utility of consciousness? Does it actually improve evolutionary fitness? In the story, the answer he explores is no: consciousness is a cognitive bottleneck that humans developed that ultimately causes our extinction against a superior non-conscious alien race
I really wish that Blindsight was made into a movie like Project Hail Mary, but unfortunately, popular audiences can't stomach demoralizing endings
There's a really cool science fiction novel - Blindsight by Peter Watts - where aliens (much creepier and more creative than Project Hail Mary's aliens) manipulate humans by moving only in intervals where humans can't consciously perceive them
Blindsight deserves so much visibility for its creativity and its biological plausibility
39
59
1,215
50,565
In a world where anyone can find bugs with a well-crafted /goal, the only impressive finds are in _bug-sparse_ code and interfaces (think vsftpd, iMessage, OpenSSH, etc).
Armadin researcher @mhskai2017 has uncovered an authenticated RCE vulnerability in Skype for Business Server (CVE-2026-70297, CVSS 8.8).
An attacker can exploit a path traversal vulnerability to write arbitrary files to the server while holding a valid token that can access any chat room with file-upload feature supported. No user interaction required. Patch immediately.
Updates are available for Skype for Business Server 2019 CU8, 2015 CU13, and Subscription Edition CU1.
Microsoft advisory: msrc.microsoft.com/update-gu…
5
7
93
9,316
IMHO Anthropic obscures names because they believe mass unrest is coming soon and employees will be targeted personally, not from poaching risks.
Crazy how Anthropic started to hide the full names of engineers on its engineering blogs... never seen this happen in tech before.
After the OpenAI espionage story I have empathy why they do it, but strange to see none the less
Clearly shows eng at an AI lab is high stakes?
53
77
2,094
107,252
It sounds like Anthropic pointed Claude at Mononoke and asked it for a customized version. It's impressive, but the software architecture and design decisions needed to scale had already been made.
lobste.rs/s/plcgps/anthropic…
4
1
22
3,652
You’d expect a lockfile to make a dependency immutable. But often it doesn’t: a hacked repository can swap in malicious code for the **same package and version**, and most package managers install it anyway.
We tested 25 package managers against a simulated malicious repository. Without a lockfile, only Go and vcpkg verified every download in our tests. With a lockfile, it's fewer than half.
blog.trailofbits.com/2026/10…
5
12
68
5,405
1 expert from @trailofbits with Codex vs 1000 big 4 AI consultants. Who would win?
2
2
44
2,481
Dan Guido retweeted
Plot twist: you were piloting real drones the whole time
Ukrainian-developed FPV drone simulator “Remote Reaper: FPV Drone” has officially left Early Access, with version 1.0 now available on Steam.
Developed in collaboration with military pilots, the simulator focuses on combat FPV drone operations and introduces several features in its full release, including armored vehicles with modeled vulnerable zones, electronic warfare effects that can disrupt video feeds and control signals during the final stage of an attack, and a customizable drone editor.
Version 1.0 also adds configurable weather conditions, custom mission editing, real-time flights with other pilots and a racing mode.
Players can use FPV hardware or a gamepad with realistic flight physics, or opt for simplified controls using a keyboard and mouse.
#RemoteReaper #FPV #FPVDrone #DroneSimulator
4
1
22
2,370
I would have ranked Immunity higher, such an injustice @daveaitel
7
15
3,239
Dan Guido retweeted
Please enjoy this short story about Annie, @alessandrogario's extra emotive cat, that resurfaced on our internal Slack today 😆
A remote @trailofbits employee had his apartment robbed while on our weekly company-wide videoconference. He had headphones on and was sitting quietly in his office, so the robbers didn’t know he was home.
4
17
Dan Guido retweeted
I notified Anthropic of these cross-session replay attacks in May, and was told that “it isn’t an attack”. I continue to be aggravated by this.
Over the last few months, we’ve made significant progress in reducing the effectiveness of distillation attacks against Claude.
In this post, we explain how: anthropic.com/research/disti…
10
19
430
23,973
Dan Guido retweeted
Tired: Two-pizza teams
Wired: Two-slice teams
@danshipper on the best team structure in the AI-era
178
200
2,154
561,308
Dan Guido retweeted
More reason not to trust what looks increasingly like an #AI cult. Having built an idol they claim is all-powerful, now they want to protect it from disrespect. Perhaps they should rename it. "Zardoz is displeased by your query!"
Anthropic has just prohibited, without explanation, "abusive behavior toward our models." But it links to an Aug. statement that it is "uncertain about the potential moral status of Claude" and wants to protect it from "distressing interactions." It's anthropomorphizing machines
2
2
8
783