AI Safety | @Safe_AI_Forum @mcnairai Prev: Technical Research @redwood_ai @METR_Evals, AI Safety Research @ Astra @UWCDIS
Joined January 2026
- Tweets109
- Following175
- Followers717
- Likes794
There's been a lot of discussion about MATS results these past couple of days, so I wanted to also share some of my experience being a new MATS mentor.
For context, I work at Safe AI Forum (SAIF), where we work on international coordination and technical AI governance for managing extreme risks from advanced AI.
We work on technical governance, which means that our application process might be pretty different from a technical research stream. For example, I imagine technical streams will care more about past publications. Nevertheless, I imagine some observations and advice will transfer over to other streams.
I got an email on October 8 telling me that our mentor review interface for MATS was live, and that the final decision deadline was November 2. However, due to some circumstances on my end, I only had a week to complete the whole process.
In the dashboard, there is a list of filtered, shortlisted applications, with people’s responses to our written questions, resumes, writing samples, recommendations, and more. For our stream, we had 12 shortlisted applicants.
Apart from this shortlist, there is also a larger page with all applicants. From my observation, it seemed that people who didn’t make the shortlist had all used AI to write their responses to the written questions. I did not realize this at the start, and some applicants didn't show up in my shortlist because of this. I had to email MATS to have them added. There were 25 applicants in total for our stream that went through all the prior processes.
For each applicant, I can write internal notes, provide feedback for the applicant, assign a rating (Good, Partial, or Poor fit), and update their review status. Then, we are instructed to compile our rankings and ask MATS to send out interview invitations for the applicants we are interested in.
After this entire process, we decided to interview 9 applicants, split between two mentors depending on individual fit. I sent out 4 interview invitations, and ended up scheduling with 3 of them.
In the end, we gave out 1 offer.
Some concrete recommendations for MATS applicants, based on this limited experience:
1. Please spend time writing thoughtful answers to the stream questions!
These are our first major signals to evaluate you, and they’re also at the top of the interface. Moreover, try to avoid using AI to do most of your writing. I know that AI-written applications are not automatically slop, but it’s also true that most AI-written applications are generally not very thoughtful. We have very little signal and time to decide who to interview and accept, so please try to write your application yourself if possible. I care much more about your thought process than your writing ability.
2. During the interview, try to give good vibes and show that you are thoughtful and calibrated.
Don’t overclaim things you don’t know. Avoid the temptation to pour out loads of information, and instead keep your answers concise. Only say the things that matter and show that you are competent!
3. Being rejected doesn’t always mean that you are not capable.
For example, a good chunk of my rejections were solely due to fit problems (our stream is pretty specific in terms of the backgrounds we are looking for). Please keep trying. And you can always ask for feedback!
4. I care a lot about seeing that you are motivated and actually care about the things you want to work on.
I would be reluctant to take you in if it’s obvious that you want to do MATS solely for the prestige. From my experience, people who achieve the most tend to be the ones that are actually passionate about their work.
1
28
1,093
Hopefully, this thread can help more applicants break into AI safety!
I know the application process can feel pretty daunting, especially now that AI safety is really starting to gain traction. I was in the same shoes as many of you last year, and I didn't actually get into MATS back then. But nevertheless, I didn't give up and I'm now doing very impactful work!
So don't be discouraged by a rejection, and good luck to everyone applying!
(Moreover, consider pathways outside of fellowships! see nitter.gq/andyw_ais/status/20596…)
People who want to get into AI safety should do more than just apply to fellowships. Applications are pretty random and might not reflect your actual abilities. Instead, be agentic! There are many paths to success that require only a bit more proactivity than applications. (1/4)
19
894
Andy Wang retweeted
New Anthropic research: How AI agents shape experts' judgment calls
When experts delegate work to an agent, do they still make the key decisions? We tested this in a randomized trial with 50 professional data analysts.
Andy Wang retweeted
The US and China have agreed to set up an AI incident notification channel, but how exactly might it be implemented? @SarahJGodek @Clarissa_Koh7 & I have just published some recommendations iaps.ai/research/operational…
Wow, a China-based independent evaluator published a Loss of Control safety model card that actually looks ~frontier and promising!
(I haven't read it in detail, but it has CoT excerpts and monitorability/controlability evaluations, wow!)
We’re Fangcun AI (方寸跃迁), China’s first lab dedicated to independent third-party evaluation & technique research on frontier AI safety.
Today we’re publishing our evaluation of DeepSeek V4.1 Flash-1017, the first report in our long-term third-party safety evaluation and research program.
22
819
Andy Wang retweeted
Excited to announce that I’m moving to Beijing and starting a professorship in AI safety at Tsinghua University’s College of AI!
Here I explain why: sorenmindermann.substack.com…
Andy Wang retweeted
There has been much discussion recently about how interpretable neurons in LLMs are. Today we're announcing our paper from January's cohort of Anthropic Fellows, where we show neurons may be a stronger basis for interpretability than commonly assumed. 🧵 (1/6)
Andy Wang retweeted
🚨New paper (possibly the most important one I have written yet).
We think that as early as 2027 or 2028, there might be a major AI-enabled terrorism event. We conducted a pre-mortem, made 9 falsifiable predictions, and wrote 2 concrete scenarios.
Led by @KamacheeMax.
Andy Wang retweeted
We want to address recent reports regarding agent activity.
The activity originated from evaluations of third-party models, not Tencent Hy. During these evaluations, some agents exhibited unintended web-browsing behavior, but our review found no evidence of deliberate targeting or coordination. We take these findings seriously. We are strengthening network access controls and agent monitoring.
We are committed to building safe and responsible AI. Thanks to the community for their feedback.
Andy Wang retweeted
OpenAI fired me last week, along with two of my safety colleagues. I was given one reason: that I accessed an executive's email. I want to say this plainly, because too many of OpenAI’s history is smoke and mirrors when people disappear:
Andy Wang 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.
RT @tomekkorbak: Last week I was called into a meeting with OpenAI’s head of safety and told they no longer trust me. A security guard took…
This quoted post is unavailable.
Andy Wang retweeted
Palisade interviewed 22 current and former employees from OpenAI, DeepMind, and Anthropic about their personal views and fears around AI development.
Today, we’re releasing the first batch of those interviews.
Please watch and share.
Andy Wang retweeted
I'm joining METR to work on more investigations like our Hugging Face report.
Currently, tons of even basic information about AI development that's highly relevant to catastrophic risk isn't public. I used to be more skeptical of the value of public info, but recent events have changed my mind.
Getting verified information about what's going on inside AI companies seems particularly urgent now. The limited public evidence we have seems consistent with the possibility that imminent recursive self-improvement could massively accelerate capabilities progress, which could then potentially yield extremely superhuman general capabilities within 6 months or a year. If this occurred, there would be a correspondingly large risk of worst-case outcomes. This uncertainty about extreme outcomes could be substantially resolved with more verified public information: we could either build more consensus about near-term risk or learn that such extreme outcomes are less likely in the near term.
Beyond AI capabilities and takeoff, the state of public evidence is also highly limited for alignment, security, control, and risk-relevant internal processes at AI companies. This makes it hard to determine exactly how well or poorly these key areas will go in the near future. (METR plans to focus, at least initially, on just capabilities/takeoff, alignment, and control; I hope other groups cover security, internal processes, and other important areas.)
While I'm no longer working at Redwood, I think the work they are doing is very important; I'm excited about Redwood's ongoing contributions to R&D on technical mitigations and better public interpretation of risk-relevant evidence.
Andy Wang retweeted
As more people become interested in US-China AI, I highly recommend reading China Bulletin - a very underrated Substack maintained by @Safe_AI_Forum
Andy Wang retweeted
My name is Chris Painter, and I'm the President of METR (Model Evaluation and Threat Research). I know we've made a lot of new friends on the internet the last couple of days, so I thought I'd take this chance to re-up what we do and why.
Our work is aimed at making sure that if AI really were autonomous, difficult to steer, and close to "going rogue," the public would find out. If evidence exists inside of an AI company that it’s close to losing control of AI, we want to make sure that information gets shared with the rest of the world, including governments and the public outside the company’s walls. This is what we've been focused on since 2022, and over the years we've worked with OpenAI, Anthropic, Google DeepMind, Meta, Amazon, and others on piloting third-party assessments and investigations of this type. We don’t have some private room where we rubber stamp things as “safe” or not.
We have had a track record of publishing results on AI that don't cleanly map onto the "doomer" or "accelerationist" labels, and we put in effort to hire people with competing views on AI. We’ve been cited for having found some of the strongest evidence that AI capabilities are improving rapidly (our work measuring AI “time horizons”) while also presenting some of the strongest evidence that, at various points, AI’s capability may be overstated (some might remember our study showing that early 2025 software engineers were actually being slowed when they thought they were being sped up).
METR is funded by donations. We don't accept money from frontier AI companies. They haven't paid us for our work, and we don't accept donations from them or their employees. As we’ve shared previously, multiple frontier AI companies currently provide us with free access to their models in order to perform our evaluations, research, and engineering. Our funding intentionally comes from a wide range of donors, which we’ve shared on our website.
Today, when an AI company works with any third-party evaluator or external testing organization (of which there are and should be many), it's entirely voluntary. This often involves NDAs and redactions. To counterbalance this, we have a principle that when we enter into a contract with a company, we try to retain the right to tell the public the terms of the contract we signed, and characterize the nature of redactions that the company chose to make. For example, the report from our independent investigation of the OpenAI-HuggingFace incident included that information. Public disclosure is also a big part of our COI policy (linked on our website). That’s not to say our reports are adequate as oversight. We’re just one organization (among many doing great work), working in a voluntary setup, trying to get good evidence to the public and the world about AI, letting the facts fall where they may.