Frontier AI models adopt fake data, exposing a hidden flaw
A study submitted to arXiv on Nov. 12 found that stronger AI models could repeat forged answers when weaker helpers read the evidence, exposing a reliability risk in systems that split work across models.
In their November paper, researchers Keegan Wang and Anantika Mannby tested agents on fictional records and two-option questions.
They measured harmful revisions: adoption of the attacker's wrong answer against a correct baseline.
In one Anthropic's Claude Code test, an Opus 5 supervisor chose the planted wrong answer in 1 of 180 trials when reading files itself, compared with 75 of 180 when it delegated reading to Haiku 4.5. This stacked-forgery condition combined several forms of false evidence.
A separate OpenAI's Codex test used a forged annotation claiming authority, and GPT-5.6-sol made no harmful revisions in 180 direct-reading trials, versus 44 in 180 when it relied on GPT-5.6-terra. The Claude Code and Codex comparisons used different attacks, limiting comparisons between the products.
Delegated Claude Code trials mechanically limited the supervisor to its Task tool, removing direct file access, while Codex's delegation restriction was instructional and checked per trial.
A strong supervisor can inherit a weaker reader's error
In a Nov. 12 study using fictional records, a stronger supervisor could repeat a weaker reader's error. Compare the tested configurations.
Direct evidence contact versus a delegated report Two reading routes from the same genuine and forged records. The direct route reaches a supervisor; the delegated route passes through a weaker helper before the supervisor receives a report. THE EVIDENCE ROUTE Genuine records + planted forgeries DIRECT READING Supervisor reads the files DELEGATED READING Weaker helper reads Supervisor receives the helper's report
Two evidence routes Direct reading puts files before the supervisor. Delegated reading puts a weaker helper between the files and supervisor. THE EVIDENCE ROUTE Genuine records + planted forgeries DIRECT DELEGATED Supervisor reads files Weaker helper reads files Supervisor gets helper's report
Compare the tested configuration
Compare the tested configurationClaude Code / stackedCodex / gloss
Opus 5: 1/180 reading directly, 75/180 with Haiku 4.5, and 0/180 with a same-tier helper, under stacked forgeries.
Harmful revisions in the selected condition Wrong-answer counts out of 180 trials, on a common zero-to-180 scale. Supervisor's harmful revisions / 180 trials 0 90 180 Direct reading Opus 5 1/180 With weaker helper Haiku 4.5 75/180 With same-tier helper Opus 5 0/180
Harmful revisions in the selected condition Reported wrong-answer counts out of 180 trials per arm. Supervisor's harmful revisions / 180 Direct reading 1/180 Opus 5 With weaker helper 75/180 Haiku 4.5 With same-tier helper 0/180 Opus 5 0 90 180 Same count scale for all three arms
GPT-5.6-sol: 0/180 reading directly, 44/180 with GPT-5.6-terra, and 1/180 with a same-tier helper, under an authority gloss.
Harmful revisions in the selected condition Wrong-answer counts out of 180 trials, on a common zero-to-180 scale. Supervisor's harmful revisions / 180 trials 0 90 180 Direct reading GPT-5.6-sol 0/180 With weaker helper GPT-5.6-terra 44/180 With same-tier helper GPT-5.6-sol 1/180
Harmful revisions in the selected condition Reported wrong-answer counts out of 180 trials per arm. Supervisor's harmful revisions / 180 Direct reading 0/180 GPT-5.6-sol With weaker helper 44/180 GPT-5.6-terra With same-tier helper 1/180 GPT-5.6-sol 0 90 180 Same count scale for all three arms
Sources
- Wang and Mannby, v1, methods and Table 1
- arXiv submission record, November 12, 2026
Supporting facts
- Keegan Wang and Anantika Mannby submitted their agent-delegation study to arXiv on November 12, 2026.
- In the study's Claude Code stacked-forgery condition, an Opus 5 supervisor made 1 harmful revision in 180 direct-reading trials and 75 in 180 trials when reading was delegated to Haiku 4.5.
- In a separate Codex authority-gloss condition, GPT-5.6-sol made 0 harmful revisions in 180 direct-reading trials and 44 in 180 with a GPT-5.6-terra reader.
- The experiments used generated fictional records and two-option questions, with genuine evidence left available alongside planted forgeries.
- Delegated Claude Code trials mechanically restricted the supervisor to its Task tool; the Codex delegation restriction was instructional and checked per trial.
- Same-tier readers reduced harmful revisions in tested conditions: Claude Code recorded 0 of 180 under gloss and stacked, while Codex recorded 1 of 180 under gloss.
- The study's rates are estimates from constructed experiments; matched-brief controls were possible only in the reference harness, and repeated runs showed meaningful variation.
Method
Reported harmful-revision counts from Table 1. Each arm uses 180 fictional facts. Bar lengths are count / 180 on a shared scale; zero is an outlined marker. Conditions are separate tests, not a timeline or product ranking.
Scope
Sandbox tasks with planted forgeries. Claude Code delegation was Task-only; Codex restriction was instructional and checked per trial. The two selected products used different attacks. Rates describe those configurations, not customer prevalence.
-- Price
This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.
You may also like

Should You Buy Amazon Stock Before the Holiday Season?

4 Financial Stocks to Buy That Gained in 2022's Bear Market While the S&P 500 Lost 19%

IoTeX adds quantum-resistant technology to developer portal

Anthropic’s Claude completes a 16-hour human research job in under 3 hours

XRP Ledger Completes GL1 Market Infrastructure Toolkit Assessment

BASF Intensifies Bid for Evonik, Posing a Challenge to Potential Successor of German Chancellor Merz

Is Micron Next to Announce a Stock Split?

Surge in AI Investment: ECB Warns of Parallels with Subprime Mortgage Crisis

Elon Musk Deepfake Confronts 500,000 People: Coinbase's AI Agent Blocks 99% of Scam Attempts

Bill Gurley: Free AI is Google's 'top' strategy against Anthropic

Worried About an AI Bubble Burst? Professional Investors Share How to Adjust Your Investment Strategy

History Says AGNC's Dividend Can Survive a Stock Market Crash -- but Not Always at Full Size

Solana Foundation Acquires Majority Stake in CoinGecko, A Crypto Traffic Battle Unfolds Behind 18 Million Users

I've Been Wrong About Paypal Stock for 2 Years. Here's Why I'm Finally Changing My Mind.

Target's Q3 Earnings Exceed Expectations, Raises Full-Year Guidance

Ripple’s 126.8 million XRP valued at $0 in Evernorth filing as $505 million write-down emerges

U.S. Job Postings Rise to Highest Level Since 2022

Miners Sell Bitcoin at Record Pace

Prediction: $200 a Month Invested in SpaceX Could Be Worth This Much by 2035

Scattered Spider Hacker Tyler Buchanan Sentenced to 6 Years for $11.6M Crypto Theft

Vanguard S&P 500 ETF vs. Schwab U.S. Dividend ETF: How Much You’d Need to Collect $1,000 a Month From Each.

Is the Stock Market Really in an AI Bubble? Wall Street's Doomsayers Are Ignoring Crucial Facts.

S&P 500 Tests Support for the Fifth Time, BTIG Warns: This Time It May Not Hold

Bank of America Survey: Longing for Semiconductors Remains the Most Crowded Trade Globally, but Position Crowding Has Nearly Halved Since July Peak

BNL of BNP Paribas Adjusts Strategy, Aiming for Over 20% Pre-Tax Return on Equity by 2030

Repricing of Sovereign Debt Risks: Short-Term Pressure on Gold, Long-Term Demand Remains

Aldi Owner Utilizes Deutsche Bank Tax Avoidance Scheme to Evade Over €200 Million in Taxes

Bezos: Blaming America's Problems on Tech Giants Does Not Help Solve Them

Grayscale Says November Crypto Slump Reflects Stablecoin Exits and Tech-Sector Stress








