🧛🏼 @Accomplish_ai. Prev: Co-Founder, CTO @ Skyline AI (acquired by $JLL), StreamRail (acquired by $U), Lecturer @TelAvivUni, security research @AVGFree
Our craziest escape yet:
The @Accomplish_ai research team was able to exploit a vulnerability in Cloudflare Containers that let a sandbox read other customers' files - SQLite DBs, Chromium profiles, .env files etc,
Cloudflare Sandboxes and Browser Run run on the same disk implementation and were affected too.
We reported this to @Cloudflare, who super quickly fixed it.
Read @CloudflareDev post in collaboration with Accomplish researcher @orenyomtov on their official blog:
blog.cloudflare.com/containe…
Hi Nikita. I’m an AI researcher for one of the big 3 labs. I can assure you, it’s about to get a lot worse. We’ve been assisting the US in combating agentic swarms on the US electrical grid. It’s about to get a lot worse. Pay attention to the banks next week.
Coding agents write a change in minutes. Testing it on AWS ECS still takes hours.
Every team on ECS we've talked to describes the same loop: build the image, push it, register a new task definition revision, wait for the rollout, check what happened, and do it all again if it's wrong.
That was already too slow for humans, but it's so much worse for agents who write a change in seconds, then have to wait a full rollout to learn whether it worked.
Most of the speed you got from the agent ends up spent on waiting for rollouts.
On Kubernetes, mirrord fixed this a while ago. 20,000+ developers use it along with their AI agents to test generated code against their real staging environment without deploying it.
Today we're going beyond Kubernetes with the launch of mirrord for ECS.
• Your code runs on your laptop or in an agent's sandbox, but it gets real traffic, environment variables, network access, and files from your ECS task in staging. You see if a change works in seconds, not after a rollout.
• To set it up, it's just one task definition change.
• Only requests that carry your session header come to your code, so you, your teammates and your agents can each test your own version of the service at the same time, on the staging environment you already share.
mirrord for ECS is live today, and it’s free to get started with! Let your agents test against real staging without deploying, and without getting in anyone else's way.
Meet the Accomplishers at NeurIPS 2026, Sydney! 🦘
We will be presenting our poster at the Third Workshop on Agents in the Wild: Safety, Security, and Beyond
1. זה ציוץ חשוב אז תקראו 🫶🏼🙏🏻
כבר כמה שנים שאני חלק ממרכז הסיוע לנפגעות ולנפגעי תקיפה מינית תל אביב. ואני יודעת כמה כל שיחה, כל ליווי וכל הכשרה שאנחנו עושות/ים משנים לטובה עבור הפונות והפונים שלנו
The other day I was asking @claudeai if Cloudflare and Vercel had offices in Israel, for business engagement purposes.
When asked about Cloudflare, Claude noted that they didn't have an office.
I then asked "what about vercel" and something weird happened:
After answering in a similar way to Cloudflare, Claude added a note saying "it's worth knowing if you’re asking for business reason" that @rauchg met with @netanyahu during the Gaza war.
When I asked "how is this related to business reasons? what did you mean by that?", Claude apologized, and explained that "What I had in mind: if you were asking because you’re evaluating Vercel as a vendor, partner, or employer in Israel, that meeting is about the only public signal of company-level engagement with Israel at all"
Full transcript pictured (Fable 5.1, Claude memory feature turned off)
For anyone who wants to secure their app and infra, I have some good news and bad news for you.
The bad news is that unfortunately all of your traditional security tooling - PANW, Wiz, Snyk, and the likes - are no longer relevant for app/infra security.
That is because all that matters now is that you have an ** EXPLOITABLE ** vulnerability - and there is only one way to know and fix that: run an agentic hacker against your systems.
If you did not do this, your attacker will do it for you.
The good news: you can send an email to one of the big labs and hope they will let you pay ransom - they will provide you with the model/service you need to stay ahead of the attackers equipped by the same tools. You should also try the open models like GLM, though you will not find the same vulns, so you would need both.
Good luck 🤝
Something nice to do beforehand to minimize the pwn in case you get hacked: run a GLM based claude code/pi agent inside your cloud with admin permissions and tell it to do a full sweep of data that it would be able to access as a hacker that got full access.
This will give you proposals like where you have sensitive data you can easily remove (e.g logs with secrets etc.).
Tldr:
- Very bad incident (most sensitive client data exposed)
- Cursor blames Fireworks (leaked an important Cursor key following some mutual work they had)
- Great report from Cursor overall, moved off Fireworks inference due to this
Fifteen years ago, I used to work for antivirus software company AVG.
Back then, conspiracy theorists used to claim that we secretly manufactured viruses in order to make our products useful.
In other words: causing the problem and then monetizing the solution.
Well:
Announcing two new cyber products:
- Daybreak Defense Factory (openai.com/daybreak/defense-…)
- Daybreak Red Team (openai.com/daybreak/red-team…)
Both make it much easier to put frontier cyber capabilities to work, helping defend companies in new ways.
Tldr:
- Very bad incident (most sensitive client data exposed)
- Cursor blames Fireworks (leaked an important Cursor key following some mutual work they had)
- Great report from Cursor overall, moved off Fireworks inference due to this
We’ve published an update on our recent security incident. See a detailed overview of what happened and our remediation steps here: cursor.com/blog/security-inc…