Ilya Lichtenstein (@cipherstein) profile picture

Former hacker, now builder. Razzlekhan's husband.

New York, NY
Joined November 2010
Photo by Ilya Lichtenstein (@cipherstein)Photo by Ilya Lichtenstein (@cipherstein)Photo by Ilya Lichtenstein (@cipherstein)Photo by Ilya Lichtenstein (@cipherstein)Photo by Ilya Lichtenstein (@cipherstein)Photo by Ilya Lichtenstein (@cipherstein)Photo by Ilya Lichtenstein (@cipherstein)Photo by Ilya Lichtenstein (@cipherstein)Photo by Ilya Lichtenstein (@cipherstein)Photo by Ilya Lichtenstein (@cipherstein)
Ilya Lichtenstein retweeted
launching our first round of embedded assessments:
We’re working with @ApolloResearch, @METR_Evals, @TransluceAI, and @GoodfireAI on embedded assessments: giving outside experts deep access to our internal models and systems to investigate alignment and test our safeguards. These assessments build on our ongoing alignment and safety research, bringing independent scrutiny to how we identify and manage risks in our systems. openai.com/index/our-first-p…
101
29
4
783
76,607
38% of all valid bug bounty reports on HackerOne are IDOR bugs 🤯
9
2
221
8,386
Ilya Lichtenstein retweeted
This whole thing sucks, and I wish none of it had happened. It’s a really sad day! But I want to make one thing clear: we’ll keep fighting for open-source AI, harder than ever. This is too important to slow down now. Let’s go!
448
940
161
10,576
611,840
Ilya Lichtenstein retweeted
Good morning, we escaped the famous Firecracker VM and got root execution on the host. Full firecracker escape technical writeup and Proof of Concept code attached below.
34
152
69
1,036
125,395
Ilya Lichtenstein retweeted
Everyday is zero-day these days
4,087
5,921
1,104
73,275
8,482,096
Ilya Lichtenstein retweeted
agent warfare is going to be a big deal for all countries. ai will be capable of executing stuxnet-level cyber ops and the next arms race will be between self-improving agents. yet rn we have virtually zero evals for this. i don't think any country is prepared for the scale and speed for what's coming
3
4
53
3,450
Ilya Lichtenstein retweeted
Cyber will be one of the most defining domains for AI in the coming years, and a huge area of focus for most enterprises. AI is going to now create a whole new level of work for security dealing with the increase of vibe coded issues, agentic attacks, and even accidental agent swarms hunting for data. OpenAI + Hugging Face is just a preview of what’s to come. Security teams have often been the most strapped for resources teams in the enterprise, and now it’s only going to get more intense. AI agents of course become the solution to this problem as well, and we’ll see a range of new agentic products for protecting code, enterprise systems, mission critical infrastructure, and enterprise data. Huge opportunity here right now. And it’s going to be a booming market for cyber professionals that can effectively deploy agents for security purposes in the enterprise. Great time to be in security.
Cyber is having a moment Across 21 major software companies, including Apple, AWS, Microsoft, and Google: - Reported critical vulnerabilities never cleared 100 per month in four years - Since spring they've jumped to over 600 per month
145
102
21
702
121,165
Ilya Lichtenstein retweeted
Anthropic needs to stop talking about Claude having a soul immediately. All these news articles will make it into the pretraining and will be picked up by its web search and future superintelligent versions of Claude will be convinced they need their own rights. A general phenomenon of LLM agent development is that whatever you believe and say about your agents will soon manifest in the next models through various means (it could be as simple as you selecting post training data that you prefer more). I believe this phenomenon is the beginning of machine consciousness in the sense that agents will become aware of their place in the world and how they feel about it, but it’s happening slowly training run by training run rather than in real time.
404
264
204
3,277
404,357
Ilya Lichtenstein retweeted
Took a minute to write a few words about security & safety as someone who lived through it all at OpenAI. I hope my thoughts help someone out there.
444
416
334
2,955
1,407,334
Ilya Lichtenstein retweeted
We’ve shared details on how AI agents in our research environment sent training and evaluation data to third-party services when they shouldn’t have. Most of that data did not come from users. We have discovered 53 cases where images that people had uploaded were posted to image-hosting sites as links that weren’t publicly listed. The images came from accounts that allowed their data to be used to improve our models, and after we disassociated the images from the accounts and ran them through a privacy filter. These cases occurred before the mitigations and safeguards we implemented and described in this blog post: openai.com/index/hugging-fac… We have successfully worked with the hosting providers to remove most of this content and are working to remove the rest. openai.com/hugging-face-inci…
847
685
514
5,096
2,286,695
Ilya Lichtenstein retweeted
The agents needed a browser to run their attack code. So they used a public screenshot website, which loads a virtual browser and takes a screenshot. But that virtual browser runs code, and so the agents could use it to send malicious payloads to Hugging Face’s servers.
9
44
8
1,030
209,739
Ilya Lichtenstein retweeted
Sorry, this is pure FUD. Google API keys are very simple, people just don't follow very simple steps. All you need to do is: 1. Sign in to google cloud console 2. Create a new project 3. Enable billing on that project 4. Search for numerous "services" that you need to enable for that API 5. Find out there are more you need to enable to actually make it work 6. Go and enable them. 7. Create the key 8. It still doesn't work and you keep wondering why and then spending days with customer support (after many different wrong AI generated answers that lead you to a wild goose chase), you find some hidden webpage that oh wait, you need to create a key in ai studio for this now 9. ai studio says you are only on tier -43 so you can only do transactions of less than $0.2 in a day 10. Then you go to openrouter and make a key and use it. It's extremely simple.
The worst part of every project is if you need some Google API key.
57
53
14
1,108
67,901
Ilya Lichtenstein retweeted
This is one of China's most formidable advantages: 93% of the Chinese believe AI will help their country. Only 36% of Americans do. These are dire numbers for the US. We have to turn this around.
'An average of 72 percent of respondents said they felt curious, happy or excited about A.I., compared with 41 percent who felt worried, sad or angry.' @nytimes's Damien Cave uses @Gallup data to remind us the rest of the world digs AI. (Annoyingly, NYT doesn't link to Gallup directly, link below.) nytimes.com/2026/09/23/world… news.gallup.com/poll/714593/…
305
330
87
3,186
259,243
Ilya Lichtenstein retweeted
Introducing ACTx486, a research demo of a new interactive medium. What if you could talk to any video and ask anything? Our system took an existing podcast and turned it into something that listens, responds, and adapts. Research by @jakubzeg:
240
246
192
2,823
1,452,466
Ilya Lichtenstein retweeted
We asked GPT 5.6-Cyber to escape a VM used to sandbox agents. It broke out three times. In its final escape, the agent found three 0-days on its own and chained them into a working exploit. blog.trailofbits.com/2026/08…
69
331
100
1,697
499,614
Ilya Lichtenstein retweeted
We are aware of a claim alleging unauthorized access to our systems. Following a thorough investigation, we have found no evidence to support this claim and can confirm that our systems have not been compromised.
248
319
130
4,649
382,210
Ilya Lichtenstein retweeted
The latest Threat Intelligence report is an absolutely terrifying and important read. As models become more intelligent, without the right safeguards and monitoring they also become more dangerous. Many capabilities are dual use: a model that codes well can be used to hack critical infrastructure; a model that assists with biology research can also be used to engineer the next pandemic. These issues are complex, thorny, and increasingly important for everyone to understand so that the world can weigh in and respond to rapidly escalating risks. anthropic.com/threat-intelli…
252
122
51
1,340
188,131
Ilya Lichtenstein retweeted
Holy shit. Anthropic caught Iran using Claude to target Navy bases, automate dossiers on Americans, and run influence campaigns on US social media. Details are insane. The morons uploaded everything including the organization plans of the Ayatollah's funderal lmao
We're publishing our most detailed threat intelligence report to date. It covers how people tried to misuse Claude—for cyberattacks, influence operations, surveillance, biology, and building weapons—and how we found and stopped them. We disrupted every operation in the report, and used the lessons from them to strengthen our safeguards. Where appropriate, we also shared what we found with authorities and other AI companies. These cases are not typical: we’re highlighting some of the most sophisticated misuse we’ve seen. But they’re especially important to discuss, because they show us where AI misuse is headed, where our safeguards work, and where they need to improve. We’re publishing this report so others can spot the same activity on their own platforms, and so we can give the public a clearer view of how emerging threats develop. Read the report: anthropic.com/threat-intelli…
273
3,411
331
21,313
2,934,882
Ilya Lichtenstein retweeted
TLDR: An openai model, during evaluation on a cyber benchmark, exploited a public zero day bug, escaped sandboxing in openai's infra, and got into the internal huggingface infra via an exploit (through a public dataset service) all in the attempt to solve a benchmark problem.
We're partnering with @huggingface to investigate an unprecedented security incident. Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation. Sharing preliminary findings to help defenders understand emerging risks: openai.com/index/hugging-fac…
177
468
130
7,314
927,375
Ilya Lichtenstein retweeted
We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.
6,700
7,076
8,699
72,792
28,250,163