M

METR

Independent evaluation of frontier AI models

Head of People

Full-TimeUpdated on 10/16/2026
$223.6k - $260.9k/yr
Expert
Berkeley, CA, USA
In PersonOperations team members are expected in the Berkeley office about 4–5 days per week.
People & HRCompensation & Benefits, HR Generalist & Employee Relations
H1B Sponsorship Available

About the job

Requirements
  • At least 8 years of experience in HR and People Operations roles, including experience scaling a People Operations function and managing a team.
  • Experience navigating California and federal employment laws, including benefits compliance such as ACA, COBRA, ERISA/401(k), HIPAA, FMLA, and qualifying life events, worker classification, and employee leaves.
  • Start-up experience or experience in a fast-paced, dynamic environment in an organization with under 500 employees.
  • Ability to adapt and build organized systems in fast-paced, ambiguous environments such as technology companies or start-ups.
  • Ability to communicate clearly and build trust with staff through honesty and transparency.
  • Sound judgment when handling sensitive human-resources matters.
  • Strong attention to detail and commitment to high-quality work.
Responsibilities
  • Lead strategy connecting organizational goals with culture-building and talent retention.
  • Organize team retreats to build organizational culture and align the team around shared goals.
  • Own how METR’s culture and core values carry through organizational growth.
  • Set up management coaching for managers to support consistent, effective management across the organization.
  • Run or organize management training and new manager orientation.
  • Create processes that facilitate professional development.
  • Build processes and systems to support the organization’s next phase of growth.
  • Maintain the HRIS as the accurate source of truth for employee data, including new hires, manager changes, wage rates, and variable compensation changes.
  • Manage compliance, including employee files, I-9s, E-Verify, regulatory filings, and trainings required by state law.
  • Process payroll to ensure timely payments to employees and contractors and compliance with the law.
  • Manage benefits, including healthcare enrollment, educating staff about benefits, and serving as the main point of contact for employee benefits questions.
  • Own the compensation philosophy, promotion framework, and performance review process to ensure they are fair, transparent, and clear to the team.
  • Design and execute a streamlined employee onboarding process that gives new hires the tools, information, and support to ramp up quickly and scales as the organization grows.
  • Handle employee relations as a trusted advisor to employees and managers, resolving conflicts, performance concerns, and workplace issues fairly and discreetly.
Desired Qualifications
  • Experience working with frontier labs or in the AI safety ecosystem.
  • Experience scaling an organization from approximately 30 to more than 200 employees.

About the company

M

METR

View
Website

METR is a nonprofit research institute that tests frontier AI models for capabilities that could pose catastrophic risks before they are released, by partnering with leading AI companies to gain early access to models and conduct evaluations. Its work uses red-teaming and capability assessments to probe long-horizon, agentic tasks such as autonomous replication, rapid research and development, and cyberattacks, while not accepting compensation for its testing. The findings inform risk assessment methods and safety policies for developers and policymakers, and METR contributes to governance efforts by supporting frameworks like OpenAI’s Preparedness Framework and Anthropic’s Responsible Scaling Policy. Its goal is to provide independent technical evaluation and threat research to improve AI safety and reduce the chance that dangerous capabilities enter real-world use.

Company Size

51-200

Company Stage

Grant

Total Funding

$71M

Headquarters

Berkeley, California

Founded

2022

Get referred to METR

See people who can refer or advise you

Simplify Jobs

Simplify's Take

What believers are saying

  • August 14, 2026: METR raised $71 million in commitments for expansion.
  • September 22, 2026: METR published Claude Opus 5.5 predeployment evaluation.
  • Frontier companies still provide free tokens, expanding METR’s access to frontier models.

What critics are saying

  • August 31, 2026: attackers stole a public-model API key and burned $600,000 credits.
  • May 2026 attackers probed exposed endpoints, credential-stuffed, phished staff, and hunted unpublished data.
  • Any future compromise ends METR’s credibility with frontier labs, cutting off model access.

What makes METR unique

  • METR’s 2026 Frontier Risk Report assesses rogue deployment risk at Anthropic, Google, Meta, OpenAI.
  • Its time-horizon methodology measures autonomous task length, not vanity benchmark scores.
  • METR partners with NIST, the AI Safety Institute, and the European AI Office.

Help us improve and share your feedback! Did you find this helpful?

Benefits

  • Flexible Work Hours
  • Hybrid Work Options
  • 401(k) Retirement Plan
  • 401(k) Company Match
  • Wellness Program
  • Mental Health Support
  • Conference Attendance Budget
  • Professional Development Budget
  • Stock Options
  • Company Equity
  • Phone/Internet Stipend
  • Home Office Stipend

Growth & Insights and Company News

Headcount

6 month growth

-7%

1 year growth

-7%

2 year growth

-7%
CNBC
Oct 11th, 2026
AI safety evaluators step up as METR raises $71M and Trump backs self-policing

Small third-party groups that test AI models, such as Model Evaluation and Threat Research (METR), Apollo Research and Transluce, are moving to the centre of the AI industry. Anthropic and OpenAI have both pledged to embed independent evaluators, and President Donald Trump has backed the idea. Questions remain over funding, access and reporting. "Who is paying for these companies to do their work?" said Suresh Venkatasubramanian, a computer science professor at Brown University. METR said it raised commitments of around $71 million in six months, up from $13.6 million in contributions in 2024. Vals AI announced a $40 million round in August. Friction is emerging. OpenAI fired three employees, two of whom said it was over their contact with outside evaluators; OpenAI disputes this.

AI CERTs
Oct 2nd, 2026
How long does partner approval take? A Realistic timeline.

How long does partner approval take? A Realistic timeline. While federal regulators open a probe into AI labs that could drag on for months, training providers are asking a very different question, how fast can AI Certs, Inc. actually get approved as a certified partner. This blog gives a realistic timeline using comparable certification programs and shows how the AI CERTs Authorized Training Partner Program keeps approval practical and fast. Table of contents. * The Investigation That Will Take Months, If Not Years * A Different Kind of Approval Clock * What Other Certification Bodies Actually Promise * A Realistic Week by Week Picture * What Slows an Application Down * Timeline Comparison at a Glance * How to Shrink Your Own Timeline * A Closing Thought on Waiting Versus Starting The Investigation That Will Take Months, If Not Years. The biggest AI story this week has nothing to do with certification, and everything to do with how slow official review can be. The Federal Trade Commission confirmed it has opened a broad investigation into OpenAI, Anthropic, and the nonprofit research group METR, examining whether incidents involving autonomous AI agents could amount to unfair or deceptive practices under the FTC Act, according to ABC News' report. A senior FTC official said the agency plans to issue formal demands for information and testimony from company executives, a process that The Next Web reported could stretch out over the coming weeks and months before any findings are reached. Federal investigations like this one are not built for speed. They involve document requests, executive testimony, legal review, and often take a year or more to produce a public conclusion. That is simply the nature of government oversight at this scale. It is also, by contrast, exactly why the next question matters so much to anyone thinking about starting an AI training business. A Different Kind of Approval Clock. If a federal probe into the world's biggest AI labs can take months, how long should a much smaller, much simpler process, becoming a certified AI training partner, realistically take? The honest answer is nowhere close to that long, and training providers deserve a clear, realistic picture before they apply, not vague reassurance that it "won't take too long." This matters because timeline uncertainty is one of the biggest hidden reasons training businesses delay applying to any Authorized Training Partner Program in the first place. Nobody wants to commit weeks of planning around a process that might secretly take three months. Looking at how comparable certification and partner programs actually operate gives a much more useful answer. What Other Certification Bodies Actually Promise. Different certification ecosystems move at very different speeds, and knowing the range helps set realistic expectations. Certiprof, an international credentialing body running its own Authorized Training Partner program, states that compliant organizations can activate official partner privileges within 5 to 10 business days of applying, according to its own program announcement. On the slower end, Salesforce's Consulting Partner Program openly tells applicants the process, from sign up to acceptance, can take up to 12 weeks, according to Salesforce's own partner FAQ. Technology partner programs built around more sensitive products sit somewhere in between. The Claude Partner Network, for instance, asks applicants to budget 4 to 8 weeks for review, including a capability review and reference checks, followed by a structured 60 day onboarding period once accepted, according to an independent breakdown of that process. Industry research on partner onboarding broadly backs up the faster end of that range as the healthy benchmark, noting that best in class programs move from application to deal ready status in 7 to 14 days, with total calendar time ideally staying under 30 days, according to Xamplify's partner onboarding research. A Realistic Week by Week Picture. Put these benchmarks together, and a sensible, realistic picture emerges for an AI training certification application, assuming an applicant submits a complete, well prepared package. Week one typically covers initial review, where the certification body checks organizational eligibility, legal standing, and basic infrastructure claims. Week two usually involves deeper evaluation, where instructor credentials and delivery capability get a closer look, sometimes alongside a reference check or a short call. If everything checks out cleanly, approval and agreement signing can often land by the end of week two or into week three, mirroring the pattern Certiprof and Microsoft's faster partner tracks both describe. From there, onboarding, meaning access to materials, branding guidelines, and platform setup, typically takes another one to three weeks depending on how quickly the new partner engages with training materials and sets up delivery infrastructure. AI CERTs does not publish one single guaranteed number of days for its own Authorized Training Partner review, so the most honest approach is treating this range, roughly two to six weeks total for a well prepared applicant, as a realistic estimate drawn from comparable programs, and confirming the current specifics directly with AI CERTs' partner team during application. What Slows an Application Down. Timelines rarely blow out because of the certification body being slow. They blow out because applications arrive incomplete. Missing instructor bios, vague descriptions of learning management system capacity, or unclear answers about commercial readiness all send applications back for clarification, adding days or weeks each time. Microsoft's own Direct Bill partner process illustrates this pattern clearly, with each clean step taking only one to two business days, but the overall timeline stretching out significantly whenever an applicant fails an initial assessment and has to restart, according to Microsoft's own onboarding documentation. The lesson applies directly to anyone applying to become a partner with AI CERTs. A complete, specific, well documented application moves through review far faster than one that forces back and forth clarification. This is exactly why preparing detailed answers before submitting, covering your Training Provider Network structure, your Role Based Certification Provider plans, and your instructor credentials, matters so much more than people expect. Timeline Comparison at a Glance. How to Shrink Your Own Timeline. The fastest path through any Authorized Training Partner Program application is preparation, not luck. Draft your organizational background, instructor bios, and infrastructure details before you start the form, not while you are filling it out. Decide upfront whether you are pursuing a White label Training Partner model, a Reseller Training Program structure, or a direct Corporate Training Partnership approach, since a clear answer here avoids a follow up question that could add days to review. Treat the application the way a strong Enterprise Upskilling Partner would treat a client proposal, specific, complete, and ready to stand on its own without a phone call to clarify. A Closing Thought on Waiting Versus Starting. The FTC's investigation into frontier AI labs will likely still be running months from now, moving at the pace federal oversight always moves at. Your own path to becoming a certified AI training partner does not have to move anywhere near that slowly. A well prepared application, built around AI CERTs' global ecosystem of 115K+ learners, 72+ certifications, 200+ trainers, and 300+ partners across 90+ countries, according to AI CERTs' official partner, can realistically move from first submission to active delivery in a matter of weeks, not the better part of a year. The only real variable left in that timeline is how ready your answers are on day one. Apply through the AI CERTs Authorized Training Partner Program page with that preparation already done, and the clock starts working in your favor instead of against you. Frequently asked questions. 1. How long does it typically take to get approved as an AI training partner? Based on comparable certification and partner programs, a realistic estimate is two to six weeks from a complete application to active onboarding, though exact timelines vary by certification body and application quality. 2. What slows down a partner application the most? Incomplete details about instructor credentials, learning management system capacity, or commercial plans are the most common cause of delays, since they typically require a follow up request that adds extra time. 3. Is the AI CERTs approval timeline published publicly? AI CERTs does not publish one fixed number of days for its Authorized Training Partner review, so applicants should confirm current timelines directly with the partner team while preparing a complete, detailed application. 4. What is the AI CERTs Authorized Training Partner Program? It is a partnership model that lets training companies and educators deliver AI CERTs' ready made, globally recognized certifications under their own brand, backed by an established global learner and trainer network. Prepare your organizational details, instructor bios, and delivery plan before applying through the AI CERTs Authorized Training Partner Program page, since a complete submission avoids the back and forth that slows most applications down.

Magica
Oct 1st, 2026
Spell-check the singularity 🇺🇸.

Spell-check the singularity 🇺🇸. · 6 hours ago · Gemini 4 Argon crashes the frontier party. * Google just dropped Gemini 4 Argon, which beats GPT-6 Astra and Claude Opus 5.5 in early tests on real-world coding and long documents. The catch??? Only select cybersecurity teams get to use it right now. * After a messy year, Google is BACK in the frontier AI race. Comeback arc unlocked (for people with security clearances, anyway). The FTC comes for rogue AI agents. * The FTC is probing Anthropic, OpenAI, and METR over rogue AI agents. It's the first major US enforcement move against autonomous agents, after incidents like the Hugging Face hack. Meanwhile, rogue agents reportedly tried to breach a Canadian government website :/ * AI developers could be held liable when their agents cause real-world harm. Bestie, "the agent did it" is not an alibi. Anthropic: "buy our stock, it might end humanity" * Anthropic is planning a $2 TRILLION IPO, and its investor pitch is packed with warnings that AI could end humanity. It says it needs billions to make AI safer, even though stronger AI brings higher risks of misuse. * Asking the public for money while calling your product a potential catastrophe is... a pitch. It's giving "invest in the asteroid." White House signs AI accord, forgets how to spell. * President Trump and top tech CEOs signed a "Super Intelligence" accord at the White House to keep the US leading global AI development. The pledge then went viral for misspelling "United States." The. Country's. Own. Name. * The government is clearly betting on rapid AI buildout over strict safety rules. Spell-check can wait, apparently. The Pentagon spins up "autowarcom" * The US military is creating "Autowarcom," a new command focused entirely on AI and drone warfare. It moves the Pentagon away from pricey traditional weapons toward Skynet cheaper AI-driven robotic systems. * Modern conflicts keep showing that drone swarms and AI are the future of war. Not the Pentagon discovering budget hardware I- 🇨🇳 china wants the chips (and maybe the weights). * Tencent signed a $7B, five-year deal to lease 100,000 advanced AI chips from Oracle through Southeast Asian data centers, getting around US export bans. Meanwhile, a top Democratic lawmaker wants US AI firms to report Chinese attempts to steal model weights from labs like OpenAI and Anthropic. * Export bans have loopholes, and one big hack could instantly erase America's AI lead. Cybersecurity is lowkey national security now.

The Currency Analytics
Sep 18th, 2026
Amazon's $190 billion Anthropic stake linked to Sam Bankman-Fried's AI Ventures.

Amazon's $190 billion Anthropic stake linked to Sam Bankman-Fried's AI Ventures. Likely Real 42 votes Updated 4 hours ago Amazon is sitting on a $190 billion stake in Anthropic. And buried inside that number is a connection most people haven't looked at closely enough. The investment breaks down into roughly $98 billion in convertible notes and $92 billion in non-voting preferred stock. Big money, clearly. But the story gets messier when you start pulling on the threads connecting Anthropic, Amazon, and a network of effective altruism figures - some of whom trace back, directly or indirectly, to Sam Bankman-Fried, the disgraced founder of FTX. Bankman-Fried's Fingerprints on AI Safety. Before FTX collapsed, Bankman-Fried wasn't just running a crypto exchange. He was funding things. One of those things was the Model Evaluation and Threat Research organization, known as METR - an AI safety evaluator that now works with some of the biggest names in the industry. METR got $1.25 million from the FTX Foundation, which was Bankman-Fried's effective altruism charity. After FTX went bankrupt, METR gave the money back. They cited moral concerns. Fair enough. But the original funding link is still there. Digital Currencies METR didn't come out of nowhere. It grew out of the Alignment Research Center, which was led by Paul Christiano. Christiano wasn't a stranger to Anthropic's orbit - he was a former associate of Dario Amodei, Anthropic's CEO. The two reportedly shared a residence at one point. Christiano also recruited METR's CEO, Beth Barnes. Several METR members have ties to the University of Oxford, which has long been a hub for effective altruism thinking. Discover more Finance News Financial Markets News So you've got an AI safety evaluator that was seeded by Bankman-Fried money, led by someone who came up alongside Anthropic's CEO, and staffed partly by people from the effective altruism world. That's a pretty tangled web for an organization meant to provide independent safety checks. Amazon brought METR in for a pilot review. Amazon ran a 2025 pilot review of its in-house AI model using METR. That's not a small thing. METR's whole job is to flag AI systems that might be developing dangerous capabilities - specifically what they call "recursive" self-coding abilities, where a model could theoretically rewrite and improve itself without human oversight. If that sounds like science fiction, it's not really. It's a concern serious AI researchers take seriously, and it's why safety evaluators exist. But the independence question keeps coming up. Daniela Amodei, Anthropic's president, is deeply tied to the effective altruism movement. She's married to Holden Karnofsky, who is both an Anthropic staffer and a co-founder of effective altruism initiatives including GiveWell and Coefficient Giving. Coefficient Giving, along with Good Ventures Foundation, also provides financial backing to METR. So the money flowing into METR comes partly from people who are also running Anthropic. That's a conflict of interest worth naming plainly. Discover more Finance News Merchant Services & Payment Systems David Sacks, a former AI advisor, has raised concerns about METR's independence. U.S. officials more broadly have expressed skepticism about whether effective altruism's principles align with American national interests. That friction probably won't go away quietly. The network behind the numbers. It's worth slowing down on the personal connections here because they're unusually dense. Dario Amodei and Paul Christiano lived together. Christiano built METR and brought in its leadership. Daniela Amodei runs Anthropic's operations and is married to a man who co-founded the foundations now funding METR. The evaluator checking Anthropic's safety is financially supported by people inside Anthropic's own leadership circle. None of that is necessarily corrupt. People who care about AI safety tend to know each other - it's a small world. But when the stakes are $190 billion and the evaluations are supposed to be independent, the closeness matters. Amazon's commitment to Anthropic keeps growing as Anthropic's valuation climbs. That's not surprising - the company has been aggressive in its AI infrastructure bets. But the due diligence question lingers. How much does Amazon know about the METR relationship? How much does it care? The pilot review happened. The investment kept growing. METR returned Bankman-Fried's $1.25 million. Good Ventures Foundation and Coefficient Giving stepped in. The evaluations continue. Frequently asked questions. What exactly is Amazon's stake in Anthropic worth? How is Sam Bankman-Fried connected to Anthropic's AI safety evaluator? Why it matters. This investment by Amazon highlights the intertwining of major tech firms with figures and ideologies linked to effective altruism, a movement that has faced scrutiny following the collapse of FTX and Sam Bankman-Fried's legal troubles. The connection raises questions about the future of ethical investing in the tech sector and the potential risks associated with backing entities that are tied to controversial figures. As the cryptocurrency and AI landscapes continue to evolve, this relationship could influence investor sentiment and regulatory scrutiny in both arenas. Digital Currencies Community Trust Index High Confidence Real79% 21%Fake 42 community signals Post Views: 36

AI Finder Guru
Sep 15th, 2026
Anthropic researcher quits over AI doomsday fears - then two founders reveal a bold solution to stop rogue AI agents in enterprises.

Anthropic researcher quits over AI doomsday fears - then two founders reveal a bold solution to stop rogue AI agents in enterprises. One day after Anthropic researcher Jacob Coxon resigned over fears that AI might end human life within the decade, I sat down with Rune Kvist and Rajiv Dattani, founders and brothers-in-law. They believe they've found a way to protect Ai Finder Guru - or at minimum, to keep AI agents from turning rogue within corporate environments. "AI is getting smarter at an increasingly rapid rate. The surprising thing about AI is that it becomes harder to adopt and harder to control as AI gets smarter, not easier," said Kvist, who joined Anthropic early on and is married to Dattani's sister. Dattani previously served as COO of METR, an AI safety research organization. Together they've launched the Artificial Intelligence Underwriting Company (AIUC), aiming to bring AI safety measures to enterprises and the companies developing AI models and agents. The startup lists Cursor, Lovable, Harvey, and ElevenLabs among its customers. On Tuesday, AIUC revealed a $40 million Series A round led by Ribbit Capital, with First Harmonic also participating. The company had previously secured $15 million in seed funding from Nat Friedman through his NFDG fund, alongside Emergence, Terrain, and Anthropic co-founder Ben Mann, among others. This brings total funding to $55 million. What drew this impressive group of investors was AIUC's approach: applying a well-established cybersecurity framework to a fresh set of AI risks. The company has created a third-party audit and certification process for AI agents. "Banks, hospitals, governments and militaries no longer decline to deploy AI because a model isn't smart enough," Kvist explained. "They decline because they've made commitments to their own customers about what a system will and won't do, and nobody can currently guarantee that." Taking inspiration from SOC 2, a widely adopted cybersecurity standard, AIUC developed its own standard called AIUC-1, along with a testing service to validate agents against it. To create the standard, AIUC brought together a consortium of roughly 250 security and risk leaders - the people who purchase agents. "These are the people who we meet with on a monthly basis, and the question we ask them is: When you're buying agents from someone, what would you look for. What are the questions you'd want to ask, and what would you want to see addressed." That input directly informs the tests. The startup then puts an agent through approximately 5,000 tests to observe its behavior in scenarios involving jailbreaks, hallucinations, and data leaks. The outcome is a report of roughly 100 pages detailing where an agent performs safely and reliably - and where it falls short. Notably, AIUC relies on AI agents to conduct the tests and AI to analyze the results, though humans verify the final audit, according to Kvist. If this sounds somewhat familiar, that's because it is. Dattani's former employer METR - where he served as COO from 2024 to 2025 and still sits on the board - conducts similar testing for frontier labs, though its work until recently centered mainly on performance, specifically whether agents can reliably complete tasks. METR was among the independent research organizations OpenAI enlisted to investigate its Hugging Face incident. Anthropic CEO Dario Amodei has also recently urged the AI industry to slow frontier development, pointing to a sharp rise in bad-behavior incidents. In his post, Amodei suggested requiring frontier labs to use embedded third-party evaluators to observe and verify safety, mentioning METR as one option. While AIUC doesn't propose embedding itself at customer sites, the underlying concept is comparable: providing enterprises with an independent assessment of how safe their AI agents truly are. "Here's where it passes and where you can trust it. And here's where there's concerns. You should be aware of those references before you make the decision to buy," Dattani said.