Carnegie Mellon University (CMU) is a private research university in Pittsburgh, Pennsylvania, formed in 1967 by the merger of the Carnegie Institute of Technology with the Mellon Institute of Industrial Research. Its School of Computer Science has one of the field's longest continuous records in artificial intelligence, beginning before the term existed, and its footprint in contemporary AI policy runs through evaluation benchmarks, red-teaming research, and the researchers and lab founders trained there.
Origins of AI research
Allen Newell joined Herbert A. Simon's research team as a doctoral student in 1955. In 1956 the two, with J. C. Shaw of the RAND Corporation, developed Logic Theorist, generally described as the first artificial-intelligence program, and presented it at the Dartmouth workshop of that summer; the term "artificial intelligence" was coined by a small group that included Newell and Simon (Source: cmu.edu). Their subsequent work produced the Information Processing Language, the General Problem Solver (1957) with its means-ends analysis, and the physical symbol systems hypothesis. Newell and Simon shared the ACM Turing Award in 1975 for their contributions to artificial intelligence and the psychology of human cognition; Simon separately received the Nobel Memorial Prize in Economic Sciences in 1978. See Herbert A. Simon.
The institutional build-out followed: a computation center in 1956, a computer science department in 1965, the first robotics institute at a US university in 1979, the Software Engineering Institute in 1984 under federal funding, and in 1988 the first college in the world devoted solely to computer science (Source: ai.cmu.edu). Newell's later work produced the Soar cognitive architecture and the unified theory of cognition published in 1990; CMU established three of the four cognitive architectures used during the 1980s to model human perception, reasoning, learning, and language. The Human-Computer Interaction Institute followed in 1994.
The university was among the laboratories DARPA (Defense Advanced Research Projects Agency) funded in the 1960s and 1970s, alongside Stanford's SAIL and MIT — the funding pattern that shaped early US AI capacity.
Machine learning and current research
Tom Mitchell joined the faculty in 1986, wrote the field's standard early textbook Machine Learning (McGraw Hill, 1997), and became the first chair of CMU's Machine Learning Department. Cosma Rohilla Shalizi has been in the Department of Statistics & Data Science since 2005, with an affiliation to the Machine Learning Department; he is a co-author with Henry Farrell of AI as Social Technology (Knight Columbia, May 2026), which the wiki carries at AI as Social Technology (Farrell + Shalizi, Knight Columbia, May 11 2026).
Two current CMU research outputs feed directly into evaluation and governance debates. ExploitBench, a public benchmark introduced by Seunghyun Lee and CMU professor David Brumley in arXiv:2605.14153, measures a model's ability to progress along the software exploitation ladder against 41 vulnerabilities in Chrome's V8 engine. It decomposes exploitation into 16 flags across five tiers — coverage, bug triggering, engine-specific primitives inside the V8 heap sandbox, general-purpose primitives such as arbitrary read and write outside it, and finally control-flow hijack and arbitrary code execution — each verified by a deterministic oracle rather than a language-model judge (ExploitBench: A Capability Ladder Benchmark for LLM Cybersecurity Agents). It is one of the instruments used in the joint UK AISI / NIST CAISI cyber assessment recorded at UK AISI / CAISI Preliminary Assessment of Kimi K3's Cyber Capabilities (July 2026). A separate CMU paper, "Silver Bullet or Security Theater," examines organizational red-teaming practice and reports widely differing approaches, a tendency for surface-level "easy wins" to dominate, time constraints shaping outcomes, and white-box methods such as the Fast Gradient Sign Method requiring model access most red-teamers do not have.
CMU co-authors also appear on the Schema harness posted July 15, 2026 with Impossible Research and UC Berkeley collaborators, which self-reported 98.98% on the ARC-AGI-3 public set against an official leaderboard best of 13.33% at the time — figures that are not ARC Prize-verified, and which the wiki reads as measurements of an agent system rather than of a model.
Researchers and founders trained at CMU
The university's clearest current policy footprint is through people. Jakub Pachocki, chief scientist of OpenAI since May 2024, took his doctorate at CMU under Gary Miller. Yang Zhilin, founder and chief executive of Moonshot AI, researched at CMU and Tsinghua before founding the company in 2023. Eric Horvitz's historical review of the field, "Scientific Progress in Artificial Intelligence", is co-authored with Mitchell.
Relationships
- related: Herbert A. Simon, Tom Mitchell, Cosma Rohilla Shalizi — faculty
- related: Jakub Pachocki, Yang Zhilin — doctoral and research alumni now leading frontier-lab research
- related: DARPA (Defense Advanced Research Projects Agency) — early and continuing federal funder of CMU AI research
- supports: ExploitBench: A Capability Ladder Benchmark for LLM Cybersecurity Agents — benchmark paper co-authored by CMU professor David Brumley
- supports: UK AISI / CAISI Preliminary Assessment of Kimi K3's Cyber Capabilities (July 2026) — ExploitBench, a CMU benchmark, is an instrument in the assessment
- related: ARC-AGI — CMU co-authors on the Schema harness result
- related: AI as Social Technology (Farrell + Shalizi, Knight Columbia, May 11 2026), Scientific Progress in Artificial Intelligence: History, Status, and Futures — Eric Horvitz and Tom M. Mitchell (2024) — publications with CMU authors
- related: Skild AI — founded by two Robotics Institute professors, Deepak Pathak and Abhinav Gupta, who remain on the CMU faculty
- related: U.S. Army (AI Deployer) — the Army's Artificial Intelligence Integration Center was built in Pittsburgh on what its own description calls "initial successes at Carnegie Mellon University"
- related: Johns Hopkins University — comparable distributed university footprint in AI governance
Sources
- (Source: ai.cmu.edu) — CMU's own account of its AI history (institutional milestones, 1965–1988).
- (Source: cmu.edu) — The Simon Initiative timeline (Newell 1955, Logic Theorist 1956, Turing Award 1975, Soar 1983, HCII 1994).
- (Source: en.wikipedia.org) — Newell biography (Information Processing Language, General Problem Solver, physical symbol systems hypothesis, unified theory of cognition, awards).
- (Source: library.cmu.edu) — CMU University Libraries on the Simon and Newell collections.