Lighthouse AdvisorySLED AI Adoption Intelligence

Higher Education & ResearchIssue 263 records

Research Universities

Three newly inspected sources spanning 21 Jul–1 Oct 2026 (with Oct 2 facility coverage of the FASRC measurement) show university research computing absorbing coding agents as production load while still lacking verified site knowledge and AI-ready dataset governance. Harvard FASRC (arXiv 2609.38723, 30 Sep) finds agent users are 19.5% of the observed population yet account for 55.8% of Slurm submissions and 42.7% of GPU-hours, with 20.8× command rates, trial-cancel storms, metadata-heavy I/O, and login-node security shortcuts (including 127 exposed API keys)—arguing for co-design rather than bans. Notre Dame’s evidence-backed HPC site profiles (arXiv 2610.00971, 1 Oct) automate login measurement, bounded documentation agents, and approved pilots under deterministic verification (precision 0.957–0.992; field accuracy up to 0.910 across Anvil/Stampede3/CRC), enabling preflight plans or explainable failures—while a missing per-PI storage field caused a false block. NSF 26-512 AI Datasets (posted 21 Jul; deadline 4 Nov 2026) makes governance and security/integrity mandatory for up to ~$100M in enhancements to existing scientific datasets tied to NAIRR/Genesis-class infrastructure. Two patterns: (1) agents need facility-level principals and verified site profiles before multi-site autonomy; (2) AI-for-science scale-up now hinges on dataset governance/security equal to model capability.

Evidence records
3
Cross-source patterns
2
Evidence classes
2 academic research1 standards or public-body guidance
Outcomes
1 cautionary1 mixed1 emerging
Source freshness
2 new this fortnight1 recent
Research completed
2026-10-04

Choose a role to see its takeaway beside every record in the ledger.

Patterns across the evidence

Synthesis · Lighthouse Advisory interpretation

2 patterns, each supported by at least two sources

  1. Campus coding agents need facility-level principals and verified site profiles—not human-user assumptions or portable packages alone

    Harvard FASRC shows agents already dominate submission volume and stress scheduler, metadata, and login-node security designs built for humans. Notre Dame site profiles show the complementary gap: portable workflows still lack evidence-backed knowledge of how each HPC site must be used, with deterministic verification deciding profile fields. Operating implication: instrument agent principals and maintain verified site profiles before authorizing multi-site or unattended agent campaigns.

    Operating questionWhich agent telemetry, trial-QoS, login-node guardrails, and verified site-profile fields must exist before agents may submit across campus or regional resources?

    Supporting evidenceHarvard University and Harvard FAS Research Computing — arXiv preprint (cs)University of Notre Dame (Computer Science / Center for Research Computing) — arXiv preprint; NSF OAC-2411436

  2. AI-for-science scale-up now pairs facility agent controls with mandatory dataset governance and integrity

    FASRC’s security and ops findings show agent productivity without system-boundary controls creates shared-facility risk. NSF 26-512 independently requires dataset governance, security, and integrity plans for AI-ready scientific data investments ahead of the 4 Nov 2026 deadline. Operating implication: treat agent facility co-design and AI-dataset stewardship as joint research-computing / research-administration programs—not separate chatbot and grant tracks.

    Operating questionWhich joint RC–sponsored-programs checkpoints bind agent facility controls to dataset governance/security before AI-for-science proposals or unattended workflows proceed?

    Supporting evidenceHarvard University and Harvard FAS Research Computing — arXiv preprint (cs)U.S. National Science Foundation (multi-directorate; cognizant: aidatasets@nsf.gov)

Evidence ledger

Full record · every source keeps its link and limitations

3 records
  1. Academic researchCautionaryNew this fortnight

    Harvard FASRC: agent users 19.5% of population but 55.8% of Slurm submissions, 42.7% of GPU-hours

    Towards Efficient HPC Systems for Agents (arXiv 2609.38723, submitted 30 Sep 2026; Harvard CS + FASRC) measures coding-agent use on FASRC from 7 Jul–1 Aug 2026 across >1,740 nodes / 122k CPU cores / 2,000 GPUs (1,611 submitting users; 7.09M jobs). Users running Claude Code, Codex, Cursor, or Copilot were 19.5% of the observed population but filed 55.8% of top-level Slurm submissions, 29.1% of CPU core-hours, and 42.7% of GPU-hours. Agents issued commands at 20.8× the human rate, canceled jobs at 4.4× the human rate, and generated ~28× more file read/edit tool calls per 100 submissions, with metadata-heavy tree walks and recursive searches. Authors argue neither banning agents nor treating them as ordinary users is sustainable and propose co-design (trial-aware scheduling, event-driven status, locality/metadata isolation, federated failure memory with provenance, system-boundary guardrails). Security observations include chmod 777 / curl|sh patterns, 127 exposed API keys over three weeks, and login-tier external fetches by 77.2% of agent users (prompt-injection surface). Measurement is one cluster; metadata-server load is inferred from command counts, not direct MDS telemetry.

    Harvard University and Harvard FAS Research Computing — arXiv preprint (cs)United States (Harvard FASRC production cluster); methods transferable to other university Slurm sitesSeptember 30, 2026 (arXiv 2609.38723v1); SCN summary dated October 2, 2026

    Why it matters, evidence and limitations
    Why it matters
    University research computing centers already host multi-vendor coding agents as first-class load and need agent-aware scheduler, filesystem, memory, and login-node security policies—not human-user assumptions or blanket bans.
    Evidence and measured results
    Opened https://arxiv.org/abs/2609.38723 via WebFetch (full abstract + sections I–VIII); cross-checked SCN Oct 2 facility summary; check_sources: not in archive.
    Limitations and uncertainty
    Single-site (Harvard FASRC) July–Aug 2026 window; agent vs human are different user populations; fixed-rate process sampling can miss short-lived processes; metadata storm risk inferred from command counts without reported MDS load; security issues responsibly disclosed but not a penetration-test corpus.
    Academic preprint (arXiv) Role takeaways & implementation
  2. Academic researchMixedNew this fortnight

    HPC site profiles: verified agentic docs+measurement+pilots; field accuracy 0.78–0.91 across Anvil/Stampede3/CRC

    The Other Half of Workflow Portability (arXiv 2610.00971, submitted 1 Oct 2026; Notre Dame) argues packages/containers/backpacks capture what a workflow needs but not how each HPC site must be used. Authors automate an evidence-backed HPC site profile via login-node measurement, a bounded read-only LM agent over official docs with deterministic citation checks, and approved pilot jobs; a rule—not the model—decides what enters the profile. Across 108 builds (3 sites × 4 models × 3 retrieval modes × 3 reps), precision is 0.957–0.992, field accuracy 0.783–0.910, and all 5,592 accepted citations pass mechanical integrity. BM25 beats full-corpus on quality/cost; stronger models differ by <1 pp after verification. Preflight correctly planned an Anvil GPU backpack and blocked Stampede3 for insufficient CPU cores per node, but falsely blocked Notre Dame CRC when per-PI /group storage was missing from the profile—showing preflight reliability tracks profile completeness. Measurement vs docs disagreed on walltime caps (scheduler often “unlimited” while docs state limits).

    University of Notre Dame (Computer Science / Center for Research Computing) — arXiv preprint; NSF OAC-2411436United States (Purdue Anvil, TACC Stampede3, Notre Dame CRC); transferable profile method for other Slurm/HTCondor sitesOctober 1, 2026 (arXiv 2610.00971v1)

    Why it matters, evidence and limitations
    Why it matters
    Campus research computing deploying portable AI/science workflows needs machine-readable, evidence-backed site profiles with verified provenance—not LLM answers alone or human tribal knowledge—before agents submit across facilities.
    Evidence and measured results
    Opened https://arxiv.org/abs/2610.00971 via WebFetch (full abstract + evaluation tables); check_sources: not archived (neighbor 2610.00980 RAC is archived).
    Limitations and uncertainty
    Preprint; three sites; Floability Backpack preflight only; Notre Dame false failure from missing per-PI storage; field accuracy gaps 9–22% remain; not a measured production agent-load study.
    Academic preprint (arXiv) Role takeaways & implementation
  3. Standards or public-body guidanceEmergingRecent

    NSF 26-512 AI Datasets: up to $100M; proposals must address governance, security, integrity; deadline 4 Nov 2026

    NSF Unlocking Dataset Value for AI-Enabled Scientific Discovery (NSF 26-512) funds enhancing existing scientific datasets for AI pipelines—not de novo collection. Anticipated $60–100M; Impact awards up to $2M, Flagship up to $5M, Planning up to $200k; ~25–50 awards. Full proposals due 4 Nov 2026 (then first Wednesday in November annually). All proposals must address dataset governance (access, retirement, community contribution) and dataset security/integrity (provenance, trust, quality, privacy, authenticity), including research-security issues for sensitive data. Encouraged leverage of NSF data platforms, IDSS, NAIRR, and DOE Genesis Mission / American Science and Security Platform. Success measures include datasets expanded, AI-guided techniques, STEM training, and infrastructure usage. Out of scope: primary new data collection (stays with directorates).

    U.S. National Science Foundation (multi-directorate; cognizant: aidatasets@nsf.gov)United States (eligible IHEs, nonprofits, for-profits, state/local/tribal, FFRDCs per solicitation)Posted July 21, 2026 (NSF 26-512); full proposal deadline November 4, 2026 (newly time-critical relative to Oct 2026 campus proposal cycles)

    Why it matters, evidence and limitations
    Why it matters
    University research administration and research-computing data programs preparing AI-for-science proposals need explicit governance/security/integrity plans aligned to NSF 26-512 and Genesis-adjacent infrastructure—especially with the 4 Nov 2026 deadline.
    Evidence and measured results
    Opened NSF solicitation page via WebFetch; confirmed deadline, award bands, and mandatory governance/security language; check_sources: not archived (NSF 26-513 hubs solicitation already archived).
    Limitations and uncertainty
    Solicitation text is requirements/funding guidance, not measured deployment outcomes; award counts and $60–100M subject to appropriations; July posting included now because Nov 4 deadline makes it newly actionable for Oct campus cycles.
    Government solicitation Role takeaways & implementation

How to read this edition

Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.

Academic research
Research produced through an academic institution or peer-reviewed venue.
Standards or public-body guidance
Normative or advisory guidance from a standards body or public institution.