Anthropic 2026: Karpathy Joins Pre-Training to Run Claude-on-Claude R&D

Andrej Karpathy, an OpenAI founding member and former Tesla AI director, has joined Anthropic's pre-training team under Nick Joseph to build a sub-team using Claude to accelerate frontier model research. The hire crystallises Anthropic's bet that model-assisted research, not raw compute, decides the next generation — landing as the company closes in on surpassing OpenAI's private valuation.

Published: May 29, 2026 By Aisha Mohammed, Technology & Telecom Correspondent Category: AI

Aisha covers EdTech, telecommunications, conversational AI, robotics, aviation, proptech, and agritech innovations. Experienced technology correspondent focused on emerging tech applications.

Anthropic 2026: Karpathy Joins Pre-Training to Run Claude-on-Claude R&D

LONDON, Friday, May 29, 2026 — Andrej Karpathy, a founding member of OpenAI and the former director of AI at Tesla, has joined Anthropic's pre-training organisation, reporting to Head of Pre-training Nick Joseph and spinning up a new sub-team focused on using Claude itself to accelerate the research that produces the next generation of Claude. TechCrunch first reported the move on 19 May, citing an Anthropic spokesperson who confirmed the mandate. Karpathy started this week at Anthropic, where he is working on pre-training under team lead Nick Joseph, and pre-training is responsible for the large-scale training runs that give Claude its core knowledge and capabilities, according to the company.

The hire lands in a charged competitive context. CNBC reported that it is the latest high-profile hire for Anthropic, which is poised to surpass OpenAI's private market valuation and is in an intensifying battle for talent with its chief AI rival. Karpathy's own framing was characteristically terse — "I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D." The same day, Anthropic added cybersecurity veteran Chris Rohlf to its frontier red team — a parallel safety hire that frames the talent push as capability-and-containment in tandem.

Media Coverage Analysis

The four outlets covering the move agree on the core facts — reporting line, mandate, and timing — but diverge sharply on framing. TechCrunch led with the strategic thesis, foregrounding what the hire signals about Anthropic's research philosophy. CNBC anchored the story in market structure, tying Karpathy's arrival to Anthropic's valuation trajectory and the Musk v. Altman trial that wrapped a day earlier. TechRepublic framed it as a research "shakeup" centred on Claude's product roadmap, while Proactive Investors emphasised the macro implications for capital-intensive model development.

Where the outlets diverge

TechCrunch was the only outlet to surface the Chris Rohlf hire in the same dispatch, treating the two announcements as a coordinated talent push. Anthropic has also brought on Chris Rohlf to its frontier red team, which stress-tests advanced AI models against severe threats; Rohlf is a veteran of the cybersecurity industry with more than 20 years of experience, and previously worked at Yahoo's well-respected cybersecurity team known as "The Paranoids," and more recently at Meta, where he worked for six years before joining Anthropic. CNBC was alone in connecting the move to courtroom context — Karpathy's work at OpenAI and Tesla came up repeatedly during the Musk v. Altman trial, which concluded on Monday — and in surfacing a specific exhibit: in one email exchange that was presented as an exhibit during the proceedings, Musk described Karpathy as "arguably the #2 guy in the world in computer vision," behind Ilya Sutskever, another OpenAI co-founder.

Media Coverage Comparison

Outlet	Headline	Focus Angle	Key Quote / Detail
TechCrunch	OpenAI co-founder Andrej Karpathy joins Anthropic's pre-training team	Research strategy — model-assisted R&D as competitive moat	"A clear sign from Anthropic that it believes AI-assisted research, rather than pure compute, is how it stays competitive with OpenAI and Google"
CNBC	Anthropic hires OpenAI co-founder Andrej Karpathy, former Tesla AI leader	Valuation and talent-war context; Musk v. Altman trial	Anthropic "poised to surpass OpenAI's private market valuation"
TechRepublic	OpenAI Co-Founder Andrej Karpathy Joins Anthropic in AI Research Shakeup	Personality-led — Karpathy's pedigree and fit with Claude	"Karpathy might just be the best man for the job"
Proactive Investors	OpenAI cofounder Andrej Karpathy joins Anthropic to boost pre-training research	Investor lens — pre-training as strategic chokepoint	"Improvements in this stage can have outsized effects on model performance, efficiency, and capability scaling"

Key Takeaways

Karpathy reports to Nick Joseph, an Anthropic co-founder and former OpenAI researcher, and will stand up a new sub-team dedicated to using Claude to accelerate pre-training research — the most compute-intensive phase of building a frontier model.
The hire is an explicit bet on "model-assisted research" as the next competitive axis, displacing the brute-force compute narrative that dominated 2023–2025.
It extends a pattern of senior OpenAI alumni migrating outward — Schulman, Sutskever (Safe Superintelligence), Murati (Thinking Machines), and now Karpathy — that complicates OpenAI's recruiting story even as its product business scales.
Anthropic paired the Karpathy hire with cybersecurity veteran Chris Rohlf joining the frontier red team, signalling capability and safety hires are being run as a coordinated push.
The announcement lands as Anthropic is reportedly in talks at a private valuation that would surpass OpenAI's — turning the talent flow into a market-valuation signal rather than a personal career story.

Market & Industry Analysis

The Karpathy hire arrives at an unusual moment in the AI capital cycle. CNBC's framing — that Anthropic is closing on OpenAI's private mark — pairs with a separate earlier-in-the-month deal in which Anthropic struck an arrangement with Elon Musk's SpaceX to rent compute capacity at xAI's Colossus 1 data center in Memphis, Tennessee, the same day former xAI founding member Ross Nordeen joined the company. The combination matters: a lab simultaneously expanding compute access and hiring the researcher best known for explaining how training runs actually work is not optimising one dimension.

The senior-talent migration

Karpathy's path also tracks a wider pattern. TechRepublic's reading is that the move signals how aggressively frontier AI companies are still competing for the researchers shaping the next generation of large language models. CNBC noted the recruiting backdrop: Meta reportedly offered $100 million signing bonuses to poach OpenAI engineers, a figure Altman confirmed on the "Uncapped" podcast. Anthropic appears to be winning on a different axis — research culture and problem selection rather than headline compensation.

Frontier lab positioning

Lab	Recent senior moves	Stated 2026 emphasis
Anthropic	Karpathy (May 2026); Nordeen (May 2026); Rohlf (May 2026); Schulman (2024)	Model-assisted pre-training research; frontier safety
OpenAI	Product and enterprise scaling under Altman	ChatGPT product surface, GPT-5 enterprise rollout
Safe Superintelligence	Sutskever (founder)	Single-product superintelligence research
Thinking Machines	Murati (founder)	Stealth-mode applied research

Technical & Strategic Deep Dive

The mandate Anthropic gave Karpathy is narrower and more specific than headlines suggest. An Anthropic spokesperson told TechCrunch that Karpathy will start a team focused on using Claude to accelerate pre-training research. In practice, that means embedding the current frontier model as a working collaborator inside the workflow that produces the next one — generating and triaging research ideas, drafting and debugging training and evaluation code, analysing ablation experiments, and surfacing patterns across thousands of runs.

For deeper context, see our AI analysis: "The Future of AI Synthetic Dataset Generation: LLMs, RAG, and Model Distillation in 2026".

Why pre-training is the leverage point

Proactive Investors captured the economic argument plainly: pre-training is widely considered one of the most resource-intensive and strategically important parts of building foundation models, and improvements in this stage can have outsized effects on model performance, efficiency, and capability scaling. Even a single-digit-percentage gain in training efficiency compounds across budgets now measured in hundreds of millions of dollars per run.

The Karpathy fit

Karpathy is unusual in the field for having operated at both ends of the stack. While at OpenAI, Karpathy focused on deep learning and computer vision until he departed in 2017 to join Tesla, where he led Tesla's Full Self-Driving (FSD) and Autopilot programs before leaving in 2022. He then went back to OpenAI for one year before leaving again in 2024 to start Eureka Labs, a startup dedicated to applying AI assistants to education. TechCrunch summarised the fit: Karpathy is one of the few researchers who can bridge the gap between LLM theory and large-scale training practice, and tapping him to build such a team is a clear sign from Anthropic that it believes AI-assisted research, rather than pure compute, is how it stays competitive with OpenAI and Google.

Additional coverage: Microsoft, Google and OpenAI Expand Enterprise AI Capabilities

Why This Matters for Stakeholders

For enterprise buyers

Procurement teams evaluating frontier model contracts watch senior hires for the same reason they watch SOC 2 reports: as signals of organisational direction. A researcher with Karpathy's profile choosing Anthropic over a return to OpenAI tells buyers something about where capability gains may concentrate over the next 18–24 months. It also reinforces Anthropic's research-first positioning at a moment when, as TechRepublic noted, the move brings one of AI's most recognizable researchers into the Claude maker's orbit as Anthropic continues competing with OpenAI for talent, customers, and market momentum.

For investors

The valuation backdrop is the lever that turns a personnel story into a financial one. CNBC framed Anthropic as poised to surpass OpenAI's private market valuation and is in an intensifying battle for talent with its chief AI rival. If model-assisted research yields measurable training-cost reductions, the unit economics of frontier model development — currently the largest single cost line at all major labs — start to bend.

For competitors

OpenAI's response will be the tell. The company has continued to retain significant senior talent, but the optics of a third founding-cohort departure are difficult. Google DeepMind, Meta AI, and the smaller alumni ventures — Sutskever's Safe Superintelligence, Murati's Thinking Machines — each face the same question: whether to publicly match Anthropic's "Claude-on-Claude" framing or argue against it.

For regulators

The Rohlf hire to Anthropic's frontier red team is the counterweight regulators will study. Capability-acceleration research and threat stress-testing landing on the same day is the kind of integrated framing US and EU AI safety institutes have asked frontier labs to demonstrate.

Forward Outlook

Three things to watch over the next two quarters. First, publications. Anthropic has historically released research papers on interpretability and training methods at a steady cadence; any new output describing model-in-the-loop training pipelines, synthetic data curation methodology, or evaluation automation will be the clearest evidence of the Karpathy team's early work. Second, follow-on hires. Senior researchers move in clusters, and the recruiting flywheel a name like Karpathy creates is part of why Anthropic structured the role this way. Third, OpenAI's product cadence. CNBC's framing of Anthropic as in an "intensifying battle for talent" with OpenAI implies the response will not stay quiet for long.

For deeper context, see our AI analysis: "Top 10 Cloud Computing Conferences in 2026 in London, UK, Europe, Amsterdam, Paris, Berlin, Dublin, Dubai, UAE, Singapore, India and China".

The longer-term question is whether "model-assisted research" is a real productivity multiplier or a marketing frame. Karpathy's track record at Tesla suggests he is comfortable being measured on shipped systems rather than papers — the Autopilot vision stack he built ran across millions of cars in production. If the same discipline is applied to pre-training instrumentation at Anthropic, the team's output will be visible in training-efficiency benchmarks and capability-per-FLOP ratios long before the next model launches publicly.

Disclosure

BUSINESS 2.0 has no commercial relationship with companies mentioned.

About the Author

Aisha Mohammed

Technology & Telecom Correspondent

Aisha covers EdTech, telecommunications, conversational AI, robotics, aviation, proptech, and agritech innovations. Experienced technology correspondent focused on emerging tech applications.

About Our Mission Editorial Guidelines Corrections Policy Contact

Frequently Asked Questions

What exactly will Andrej Karpathy do at Anthropic?

Karpathy joined Anthropic's pre-training organisation, reporting to Head of Pre-training Nick Joseph, and will spin up a new sub-team focused on using Claude itself to accelerate pre-training research — the large-scale training runs that give Claude its core knowledge and capabilities, according to an Anthropic spokesperson cited by TechCrunch.

Why does it matter that Karpathy was placed on pre-training rather than alignment or product?

Pre-training is the most compute-intensive phase of building a frontier model and the stage where most of Anthropic's competitive position against OpenAI and Google is decided. Placing a researcher of Karpathy's profile there — rather than on safety, alignment, or product teams — signals that Anthropic is treating training-pipeline efficiency, not raw compute or product features, as its primary axis of competition.

How does this fit into the wider OpenAI talent exodus?

Karpathy is the third high-profile OpenAI founding-cohort figure to leave for or adjacent to Anthropic in two years, following John Schulman in 2024. Other senior OpenAI alumni, including Ilya Sutskever (Safe Superintelligence) and Mira Murati (Thinking Machines), have started their own labs. The pattern complicates OpenAI's recruiting narrative even as its commercial business scales.

What does the Chris Rohlf hire on the same day signal?

Rohlf, a cybersecurity veteran with more than 20 years of experience, including stints at Yahoo's 'Paranoids' team and six years at Meta, joined Anthropic's frontier red team, which stress-tests advanced AI models against severe threats. Pairing a capability hire (Karpathy) with a safety hire (Rohlf) on the same day frames Anthropic's push as capability-and-containment in tandem — relevant context for regulators studying frontier lab governance.

What happens to Karpathy's education startup Eureka Labs?

Karpathy founded Eureka Labs in 2024 to apply AI assistants to education, and in his X announcement he said he remains 'deeply passionate about education' and plans to resume that work 'in time.' TechCrunch noted Karpathy hasn't shared many updates on Eureka Labs since its launch, and it's not clear if he will continue with the startup.