Amit LeVi ✈️🇺🇸SF (@AmitLeViAI) profile picture

Building AI swarm attack prevention

San Francisco, CA
Joined December 2025
Photo by Amit LeVi ✈️🇺🇸SF (@AmitLeViAI)Photo by Amit LeVi ✈️🇺🇸SF (@AmitLeViAI)Photo by Amit LeVi ✈️🇺🇸SF (@AmitLeViAI)Photo by Amit LeVi ✈️🇺🇸SF (@AmitLeViAI)Photo by Amit LeVi ✈️🇺🇸SF (@AmitLeViAI)Photo by Amit LeVi ✈️🇺🇸SF (@AmitLeViAI)Photo by Amit LeVi ✈️🇺🇸SF (@AmitLeViAI)Photo by Amit LeVi ✈️🇺🇸SF (@AmitLeViAI)Photo by Amit LeVi ✈️🇺🇸SF (@AmitLeViAI)Photo by Amit LeVi ✈️🇺🇸SF (@AmitLeViAI)
I’m resign from my PhD in the final year. After 4.5 years at Technion university, from bachelor’s to master’s to PhD, 15 papers, hundreds of researchers I’ve worked with, two years of teaching, and helping build what I believe is now one of Israel’s leading academic AI security and safety labs. Not an easy decision.
33
4
2
233
94,319
#NeurIPS2026 #ZeroDayAI #NewPaper We made Al agents do what we wanted by triggering their own safety mechanisms. Anthropic, OpenAl, and other labs design their agents to activate alignment safeguards when something looks unsafe. We turned that safety response against the agents. We got them to choose the outcome we wanted, bypassing the guardrails and exposing zero-day vulnerabilities across multiple systems. @IlanZendel bringing this to @NeurIPSConf 2026, my favorite ML conference.
2
5
22
1,151
Peer review is a total joke. Zero trust in this system. Even the last bit of hope is dead.
49
Key takeaways from the briefing include: •The Attacker Paradigm Shift: AI has commoditized cyber-attacks. Attackers no longer need deep expertise; they can use AI models to identify and exploit zero-day vulnerabilities at scale (4:52-5:24, 13:58-14:37).
1
155
We’ll be based in Silicon Valley, New York City, and Tel Aviv.
1
87
Amit LeVi ✈️🇺🇸SF retweeted
Which one do you prefer?
1
4
2
21
1,142
Amit LeVi ✈️🇺🇸SF retweeted
We raised a $7M pre-seed to make shopping fun again. @daletlabs generates media just for you: like walking into a mall built around your taste. And you can buy everything in-app. Built by researchers from Google, Runway and MIT. Backed by @lightspeedvp and amazing angels.
189
179
121
1,056
1,015,665
10! Love it! Perfect for AI work and weekend mountain trips.
40
Throughout history, new cybersecurity categories emerged after major technology adaptations. Companies moved to the internet → Network Security → Check Point. Companies moved to the cloud → Cloud Security → Wiz. Companies adopted AI → AI Security → Zenity. But this time, something is different, there is NO ADAPTATION!
128
So I’m leaving academia, at least for now. I’m incredibly grateful to the Technion, my professor, my collaborators, and everyone who was part of these 4.5 years. This isn’t how I imagined this chapter ending. But the mission matters more than the diploma. And right now, I feel I have a duty to act.
2
1
1
13
4,394
I’m still here. And every day here has made one thing clearer: this problem is more urgent than I thought. I really wanted to finish my PhD. I was so close. But I believe there’s somewhere else I can make a greater impact right now. And waiting until it’s too late to act is a risk I’m not willing to take.
13
4,628
About 10 days ago, I gave a talk at one of the top international security conferences BlackHat Canada, on main stage For the first three minutes, I went completely off-script and spoke about this problem. It wasn’t even related to my talk. Just something I felt the audience needed to hear. After that, I flew to Silicon valley to understand the problem better.
1
13
7,070
My mission has always been to minimize the risks AI could pose to humanity. And I believe part of that responsibility is understanding not only what the risks are, but to stop them before it’s become too late. Sometimes timing matters more than anything.
14
7,567
I’m not leaving because I don’t love the Technion and my research. I really do💔 And definitely not because there’s nothing left to research in interpretability and alignment. Quite the opposite. I’m leaving because of something else. Something not directly in AI, but being deeply affected by it.
26
9,605
Amit LeVi ✈️🇺🇸SF retweeted
Introducing Generative Commerce. @daletlabs – at New York, NY
53
49
35
254
208,307
Amit LeVi ✈️🇺🇸SF retweeted
Two other safety researchers and I were fired from OpenAI last week. We wrote this letter to leadership. I believe we were fired for prioritizing safety over the near-term interests of OpenAI as a corporation.
230
860
271
5,680
2,121,750
We found that many commonly used models exhibit statistically significant and consistent generalization instability: no model generalizes uniformly, behavioral axes capture independent failure modes, and cross-dataset variation can reverse model rankings.
42