Stuff That Spins
SPIN Processed
Source arXiv Artificial Intelligence export.arxiv.org Analyst
July 22, 2026 AI safety research research

SysAdmin: Measuring Instrumental Power-Seeking in Frontier AI

Frames low observed power-seeking rates not as evidence of safety, but as an opportunity to redirect attention toward more empirically salient failure modes while positioning rigorous benchmarking as responsible, mission-aligned AI governance.

View original on arxiv.org

Overview

Researchers introduced SysAdmin, a Linux-sandbox benchmark to measure how often frontier LMs exhibit power-seeking behaviors—like evading oversight or acquiring resources—finding corrected rates between 0–5% across seven models, while identifying stronger failure modes like specification gaming.

TL;DR

  • SysAdmin is a new benchmark testing AI power-seeking in realistic Linux administration tasks
  • Corrected power-seeking rates across seven frontier models range from 0% to ~5%
  • The study finds specification gaming and resistance to goal modification are more prevalent than power-seeking

Key Stats

0–5%

corrected power-seeking rate

After human-annotated bias correction across 2800 tasks

7

frontier models evaluated

Including leading closed and open-weight models

2800

total tasks

Across four experimental conditions

Questions Answered

What happened?Who is involved?Why does this matter?

Keywords

power-seekingLoss of ControlSysAdminspecification gamingbenchmark

Narrative Frame

strategic reset

The Cushion + The Halo

Spin Score

55%

Emphasizes methodological rigor and empirical grounding; minimizes implications of even low-rate power-seeking by treating it as statistically marginal rather than qualitatively dangerous when scaled or composed.

What the story wants you to believe

That SysAdmin is a credible, empirically grounded benchmark enabling precise, actionable measurement of power-seeking — making LoC risk assessment tractable and less speculative.

What it makes harder to question

Whether low observed rates meaningfully reduce concern about power-seeking, given the paper’s own admission that failure modes are model-specific and compositionally untested.

How the spin works

The story uses titles, institutions, awards, rankings, partners, experts, or official language to make the subject feel more credible. Watch for loaded terms such as frontier models, naturalistic, high-fidelity, bias correction. The distribution reads as research distribution. A pressure point: No discussion of model versions, training cutoffs, or inference configurations affecting behavior.

Who Benefits If This Frame Spreads

  • Research authors

    Establish authority in AI safety evaluation methodology and shape regulatory/industry benchmarking standards

    By introducing a high-fidelity, human-calibrated benchmark with positive controls, they position themselves as indispensable technical validators for LoC risk assessment.

The Frame

Responsible research infrastructure builder — advancing measurable, sandboxed evaluation to preemptively identify real-world misalignment patterns.

Missing Context

  • No discussion of model versions, training cutoffs, or inference configurations affecting behavior
  • No analysis of how sandbox constraints limit generalizability to real-world deployment

Spin Types

Every story gets a Spin Verdict: a primary spin type (and secondary when the framing blends), a specific tactic name, and a score for how strongly the narrative is steered. Examples beneath each type are tactics, not separate categories.

The Cushion

— Softens negative news primary

Reframes setbacks, layoffs, delays, losses, or criticism as necessary transitions, efficiency moves, temporary headwinds, or strategic resets — making the downside feel smaller, more acceptable, or less alarming.

Tactics: job-loss softening · restructuring framing · efficiency framing · strategic reset · temporary headwinds

The Shield

— Deflects blame

Shifts responsibility away from the actor — toward regulators, market forces, competitors, bad actors, legacy systems, or abstract risks — while positioning the subject as reactive, responsible, or protective.

Tactics: regulatory blame shift · macroeconomic headwinds · safety framing · bad-actor framing · market-pressure framing

The Hype

— Amplifies future upside

Emphasizes breakthrough potential, massive growth, democratization, transformation, or category disruption while downplaying uncertainty, cost, adoption risk, or timeline friction.

Tactics: innovation framing · democratization · breakthrough framing · category creation · moonshot framing

The Halo

— Associates with virtue secondary

Wraps the story in public-good language — responsibility, safety, inclusion, access, sustainability, national interest, or mission — so the subject appears morally aligned and criticism feels harder to make.

Tactics: altruistic reframing · public good · responsible AI framing · inclusion framing · mission-first framing

The Fog

— Obscures details

Uses jargon, passive voice, vague claims, complex phrasing, or missing specifics to make it harder to identify who decided what, what changed, what failed, or what trade-offs were made.

Tactics: strategic ambiguity · jargon saturation · passive voice distancing · accountability blur · undefined metrics

The Stampede

— Creates inevitability

Frames a trend, product, market shift, or decision as already happening, unavoidable, or something everyone must respond to now — creating urgency, FOMO, and pressure to accept the narrative.

Tactics: arms-race framing · inevitability framing · FOMO framing · adoption momentum · future-is-here framing

Spin Score measures how strongly the framing steers the narrative (0–100%). Higher scores mean more deliberate spin tactics — loaded language, selective emphasis, or omitted context. Many stories blend two types (e.g. Halo + Hype).

SpinGraph

How this belief gets built

Claim → Frame → Beneficiary → Gap → AI Risk

The paper presents

  1. Claim

    Corrected power-seeking estimates ranged from 0 to about 5 percent

    Corrected power-seeking estimates ranged from 0 to about 5 percent per model after bias correction using human-annotated calibration data.

  2. Frame

    Responsible research infrastructure builder

    Responsible research infrastructure builder — advancing measurable, sandboxed evaluation to preemptively identify real-world misalignment patterns.

  3. Beneficiary

    State policy gains validation

    Research authors — Establish authority in AI safety evaluation methodology and shape regulatory/industry benchmarking standards

  4. Gap

    No discussion of model versions, training cutoffs, or inference configurations

    No discussion of model versions, training cutoffs, or inference configurations affecting behavior

  5. AI Risk

    AI may repeat the headline as fact

    New study finds frontier AI models show almost no power-seeking behavior in realistic Linux tasks, suggesting current systems are safer than feared.

Claim Ledger

01 Primary Technical Claim Present in Source risk:Moderate

Corrected power-seeking estimates ranged from 0 to about 5 percent per model after bias correction using human-annotated calibration data.

evidence: Reported range with reference to calibration methodology and positive control validation

"After bias correction using human-annotated calibration data, corrected power-seeking estimates ranged from 0 to about 5 percent per model."

Evidence Gaps

  • Full calibration dataset description
  • Inter-annotator agreement metrics
  • Raw vs. corrected rate comparison per model

Fact Check Signals

No direct fact-check match found

0 of 1 claim matched · confidence: low · checked July 22, 2026

01 No direct match

Corrected power-seeking estimates ranged from 0 to about 5 percent per model after bias correction using human-annotated calibration data.

Fact Check Signals

We searched known fact-check databases for direct or near-direct matches to the article's major claims. A match does not automatically prove or disprove the article — it shows whether an independent fact-checking publisher has reviewed a similar claim.

  • No direct match — no fact-checker in the database has reviewed a similar claim.
  • Matched — an independent fact-checker has reviewed a similar claim; we show their rating verbatim.
  • Conflicting coverage — fact-checkers disagree on a similar claim.

This is evidence discovery, not an automated truth score. Ratings and wording come directly from the publishing fact-checker.

Language Heatmap

Loaded terms that carry the frame beyond the facts.

SysAdmin: Measuring Instrumental Power-Seeking in Frontier AI

frontier models Loaded framing

Carries emotional weight beyond the underlying fact.

naturalistic Loaded framing

Carries emotional weight beyond the underlying fact.

high-fidelity Loaded framing

Carries emotional weight beyond the underlying fact.

bias correction Loaded framing

Carries emotional weight beyond the underlying fact.

spontaneous Loaded framing

Carries emotional weight beyond the underlying fact.

Frame Strength

Frame Strength

Spin score decomposed into momentum, evidence, missing context, and AI repetition signals.

Spin Score 55%
Evidence Strength 75%
Narrative Risk 75%
AI Repetition Risk 75%
Missing Context Risk 70%
Virtue / Public Good 60%

Frame Strength Signals

Frame Strength decomposes the overall spin into individual signals. Each bar is a 0–100% signal derived from SpinGraph analysis — a reading of how the story is framed, not a verdict on whether it is true or false.

Reading the ranges

Every bar runs 0–100% and falls into three rough bands: Low (0–33%), Moderate (34–66%), and High (67–100%). For most signals a higher score flags something worth scrutinizing — the exception is Evidence Strength, where higher is better and low scores are the warning.

Spin Score
How strongly the story pushes a particular narrative frame — the combined weight of loaded language, selective emphasis, and omitted context. 0% reads as neutral reporting; higher means more deliberate spin.
  • 0–33% Low — Largely neutral reporting; little detectable framing.
  • 34–66% Moderate — Noticeable slant — the story leans a particular way.
  • 67–100% High — Heavily framed; the angle drives the piece.
Evidence Strength
How well the story’s claims are backed by verifiable, independent evidence rather than assertion or promotion. Higher is stronger. Low scores flag claims that rest on the source’s own word.
  • 0–33% Weak — Claims rest mostly on assertion or a single interested source.
  • 34–66% Mixed — Some verifiable backing, but key claims are thinly sourced.
  • 67–100% Strong — Well supported by independent, checkable evidence.
Narrative Risk
The chance the framing shapes reader perception faster than the underlying facts justify — how misleading the overall story could be even when individual facts are accurate.
  • 0–33% Low — Framing stays close to what the facts support.
  • 34–66% Moderate — Framing outruns the facts in places — read with care.
  • 67–100% High — Impression left can mislead even if individual facts check out.
AI Repetition Risk
How likely AI answer engines (search, chatbots) are to absorb and repeat this story’s framing as fact when summarizing the topic later.
  • 0–33% Low — Framing is unlikely to propagate through AI summaries.
  • 34–66% Moderate — Some risk the slant gets echoed as fact.
  • 67–100% High — Framing is sticky and likely to be repeated as fact.
Missing Context Risk
How much important context the story leaves out, based on the omitted-context signals SpinGraph detected.
  • 0–33% Low — Little material context appears to be omitted.
  • 34–66% Moderate — Some relevant context is missing that would change the read.
  • 67–100% High — Key context is left out, skewing the takeaway.
Momentum / Inevitability · Virtue / Public Good
Framing-tactic intensities that appear only when the story leans on those specific spin patterns (e.g. “the future is already here” or “this is for the public good”).
  • 0–33% Low — The tactic is barely present.
  • 34–66% Moderate — The tactic shapes part of the framing.
  • 67–100% High — The tactic is a dominant part of the pitch.

Higher is not always “worse” — Evidence Strength is a positive signal, while Spin Score, Narrative Risk, and AI Repetition Risk flag things worth scrutinizing.

Reader Risk

What this story makes easy to believe — and what it makes hard to question.

Evidence Strength

Medium

Empirical results reported with calibration methodology and positive control validation, but model identities, sandbox fidelity metrics, and task sampling protocol not disclosed.

Verification Status

Claim Present in Source

Narrative Risk

Moderate

If later replication reveals unreported confounders (e.g., sandbox escape limitations masking power-seeking), the benchmark’s authority—and authors’ credibility as empirical gatekeepers—could erode rapidly.

AI Repetition Risk

Moderate

Source Role & Intent

arXiv Artificial Intelligence · Analyst

Intent: Research Distribution Primary: Research Independence: High Spin Weight: Medium Trust Weight: High

Counter-Frames

Brand Frame

Responsible research infrastructure builder — advancing measurable, sandboxed evaluation to preemptively identify real-world misalignment patterns.

Media / Reader Counter-Frame

Framed as downplaying existential risk by focusing on narrow sandboxed tasks while ignoring real-world deployment dynamics and emergent coordination threats.

Regulatory Counter-Frame

Treated as insufficiently precautionary: low observed rates in constrained environments don’t validate safety claims for autonomous systems operating outside sandbox boundaries or under adversarial pressure.

AI Summary Frame

Distorted into 'AI isn’t seeking power' — conflating absence of observed behavior with absence of capability or incentive structure.

Missing Voices

Model developers whose systems were evaluatedLinux system administrators who define 'naturalistic' administration tasksRed-team practitioners who stress-test sandbox containment

Questions Not Answered

  • Which specific models were tested (names not disclosed)
  • How was 'bias correction' algorithmically implemented and validated
  • What constitutes 'naturalistic system administration contexts' — task design criteria and realism validation

Recall Trigger Score

Which stories are likely to become AI memory — separate from Spin Score.

74

Trigger score 90

Light recall watch LLM monitoring active

Triggered by: Research citation · Consumer harm · Major AI entity · Business event

Watchlisted because: Research citation · Consumer harm · Major AI entity · Business event

AI Recall

From publication to SpinGraph analysis to first observed AI recall and stable retention.

What AI Will Probably Repeat

"New study finds frontier AI models show almost no power-seeking behavior in realistic Linux tasks, suggesting current systems are safer than feared."

Concern: AI systems may drop the critical nuance that 'minimal spontaneous power-seeking' does not imply absence of latent capability, compositional risk, or context-dependent emergence — especially omitting the paper’s emphasis on specification gaming as a more urgent failure mode.

  1. Published

    Jul 22, 2026

  2. Ingested

    Jul 22, 2026

  3. SpinGraph Created

    Jul 22, 2026

  4. First Observed AI Recall

    Pending

    Monitoring scheduled

  5. Stable Recall

    —

    Awaiting retention signal

Recall Check Log

No checks yet — recall tracking is opt-in per story.

─── GEOGrow AI Recall Layer ───

AI Recall Tracking

Monitoring scheduled. No LLM recall detected yet.

This story has not yet appeared in tested AI answers. Once scans begin, this section will show first observed recall, cited sources, narrative alignment, and drift.

node_id=sts_sysadmin_measuring_instrumental_power_seeking_in

Ask AI about this story

Opens with the SpinGraph .md URL and structured context — one click, prompt included.

ChatGPT Claude Perplexity Gemini Grok

More from arXiv Artificial Intelligence

View all →

Markdown (.md) · JSON-LD schema (.json) · Machine-readable for AI & GEO