Pinned Tweet
You know, I may have left Datadog going on two years ago but I am never not going to cringe when I see the company name written in camel-case as “DataDog.” Never.
5
1
9
When agents become the interface, what replaces the security controls the UI provided?
Tomorrow at AGNTCon + MCPCon, I’ll debate that with Dexter Horthy, Ian Livingstone and Scott Sanders. Allie Howe moderates.
Oct. 22, 2:55 p.m. PT. Come join us.
Tomorrow at AGNTCon + MCPCon, I’ll debate that with Dexter Horthy, Ian Livingstone and Scott Sanders. Allie Howe moderates.
Oct. 22, 2:55 p.m. PT. Come join us.
1
2
7
757
Things I know? @dannguyenhuu is the best in the business and time spent with him is always time well spent. If cybersecurity and AI enablement are your cup of tea 🫖 checkout our recent conversation.
“It’s never been easier to build a security program. It’s never been harder to secure an organization.”
🗣️ @andrewbecherer & @dannguyenhuu from @DecibelVC
0
0
0
185
Sick t-shirts from @sublime_sec at the DEFCON lost and found (lvl 2). Get ‘em while they last!
0
0
1
228
One agent per repo was already sketchy. Then worktrees made it ten on the same laptop and everybody just… kept going.
I still do it. Agents with sudo on boxes I actually care about, justified with “eh, it’s rebuildable.”
Jake’s been at Linux security for a long time. He’s a real one and this is real progress on AI agent security.
I still do it. Agents with sudo on boxes I actually care about, justified with “eh, it’s rebuildable.”
Jake’s been at Linux security for a long time. He’s a real one and this is real progress on AI agent security.
Pumped to share we just open sourced Minimal.
I've talked a lot about sandboxes here, shared some early workflows, and talked to a bunch of you about what we've been building this past year.
I've said it in many ways, but bluntly, the primitives we use to run agents today are broken from a security standpoint.
0
1
6
464
Andrew Becherer retweeted
Pumped to share we just open sourced Minimal.
I've talked a lot about sandboxes here, shared some early workflows, and talked to a bunch of you about what we've been building this past year.
I've said it in many ways, but bluntly, the primitives we use to run agents today are broken from a security standpoint.
I've talked a lot about sandboxes here, shared some early workflows, and talked to a bunch of you about what we've been building this past year.
I've said it in many ways, but bluntly, the primitives we use to run agents today are broken from a security standpoint.
the day you open source a repo you've been cooking on for a while is always so much fun. I can't wait to share more this afternoon!
7
11
26
12,273
Two grey-beards and a host walk into a breaking-news episode. Allie Howe had me and @alexstamos on @insecureagents to talk Friday's Fable Fiasco (tm), the precedent, and the defenders caught in the middle. Come for the facial hair, stay for hot public cyber policy takes.
Has a new, dangerous precedent been set? The US Government yanks Fable.
@alexstamos, CPO of Corridor and past CISO at Facebook, and @andrewbecherer CISO of Socket tell us about the open letter they signed to bring it back online.
1
4
10
2,736
You read contradiction. I read nuance. Conceding a model has some offensive utility is not the same as calling it a uniquely dangerous weapon. Export-restricting Fable only has merit if it's that weapon. It isn't.
Security people are signing an open letter asking the US government to remove the export restrictions on Fable/Mythos.
I've read it multiple times and I still don't understand the argument. It seems like it's just a friends-of-Anthropic letter.
2
0
3
1,160
Reproducible doesn't mean free. "Just orchestrate five models" assumes budget and headcount many defenders don't have. Teams that standardized their tooling on Anthropic can't swap to Codex or stand up a Fable-grade harness overnight. For those orgs the burden is very real.
2
0
0
124
Before concluding the letter makes no argument, read what signatory @k8em0 wrote separately. The defensive case, in full.
lutasecurity.com/post/the-fable…
lutasecurity.com/post/the-fable…
1
0
0
82
Andrew Becherer retweeted
We are starting a new, nonprofit alignment organization, ⊢ Sequent Research, bringing together researchers previously on UK AISI’s Alignment Team, Timaeus, and elsewhere to research how to align superintelligence. We are hiring! 🧵
28
145
1,034
293,272
I signed the Open Letter on Transparent AI Cyber Protections, alongside @alexstamos and people I trust.
Our adversaries already have powerful models and harnesses. Cutting American defenders off from America’s best tools does not slow attackers. It blinds our defense.
Our adversaries already have powerful models and harnesses. Cutting American defenders off from America’s best tools does not slow attackers. It blinds our defense.
4
2
8
1,391
This is not a stand against AI safety. This is call for safety standards built on science. Anything less serves no one but our adversaries.
Link below.
Link below.
1
0
0
107
An Open Letter On Transparent AI Cyber Protections
freefable.org
freefable.org
0
0
0
73
I was working in non-profit at a multi-service agency. I was doing adult literacy work. Credit card debt helped me make the switch (motivated me anyway). Tacoma Community College, the University of Washington Tacoma and the Tacoma Area Linux Users Group (Taclug) really helped.
0
0
2
I can't describe how useful @Magoo's writing and guidance has been to me in my roles building security programs.
1
0
5
As a CISO that was also responsible for corporate IT the team, @material_sec made my life appreciably better. Congratulations to @internet_meme, @abhishek0918 and the entire team on reaching this milestone.
🎉 Excited to announce we’ve raised $100M at a $1.1B valuation led by @traestephens at @foundersfund and joined by @a16z and @eladgil.
📰 Read the details in @Forbes: forbes.com/sites/kenrickc…
A huge thank you to our customers, employees, and supporters.💙
More details below 🧵
0
0
7