From the Public Safety edition of September 17, 2026
Reuters: AI-error court filings keep surging three years after first U.S. sanctions
Reuters · Courts / legal profession reporting · United States (state and federal courts)
- Publisher
- Reuters
- Original publication
- September 17, 2026
- Source retrieved
- 2026-09-18
What happened
On 2026-09-17 Reuters reports generative-AI errors continuing to appear in U.S. court filings despite three years of sanctions, with recent examples spanning a New Mexico murder appeal citing fabricated police testimony, an Oklahoma judicial order with fake citations, and a California State Farm counsel fine for erroneous citations across seven filings. A researcher-tracked count cited in the piece puts hallucinations in at least 1,395 state and federal cases. Judge Castel argues awareness has not produced enough deterrence and that stronger sanctions may be needed. Errors are also described as reaching court opinions, expert testimony, and prosecutors' briefs—not only private civil practice.
Why it matters
Directly relevant to state/local courts, prosecutors, and defense counsel: shows that ethics rules and existing sanctions have not stopped GenAI hallucination risk in criminal and civil filings, so SLED buyers need verification workflows, disclosure policies, and bar/bench training—not tooling alone.
Evidence and measured results
Opened original HTTPS article via WebFetch on 2026-09-18; recorded quantitative claim (1,395 cases), named recent incidents, and Castel quote on need for greater deterrent sanctions.
Limitations and uncertainty
Count of 1,395 depends on an external researcher's tracker linked by Reuters; Reuters does not publish a full reproducible methodology in-article. Individual case facts summarized from court events and attorney statements.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-09-18; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Attorneys and chambers keep filing GenAI-generated fake citations/testimony despite years of sanctions, creating professional and case-integrity risk.
- Stakeholders
- Presiding/administrative judges
- Clerk of court / e-filing
- District attorney / county counsel
- Public defender leadership
- State bar ethics / risk counsel
- Discovery questions
- What written GenAI filing policy exists today, and is verification attested on e-file?
- How are consumer chatbots blocked or logged on justice endpoints?
- What was the last local sanction or rebuke involving AI citations?
- Who owns spot-audits of filed briefs for hallucinated authority?
- Value hypothesis
- Bounded value is reducing unverified GenAI text reaching the docket through enterprise tools plus mandatory human cite-check—not guaranteeing zero hallucinations.
- Potential engagement
- 90-day policy-and-controls assessment: map current GenAI use, design attestation/logging, pilot cite-check workflow on one practice group.
- Unsupported claims
- Guaranteed elimination of hallucinations
- That sanctions alone have already solved the problem
- Product-specific accuracy without independent validation
Pre-sales engineering
Role takeaway
- Fit
- Conditional fit where CMS/e-filing can insert a human attestation gate and retain prompt/response artifacts; poor fit for unconstrained consumer GenAI on justice desktops.
- Architecture and integration
- Drafting aid → cite-check service → attorney attestation → e-filing; immutable logs of prompts/outputs separate from official filings.
- Prerequisites
- Written GenAI use policy
- Enterprise AI endpoint with DLP
- Audit storage retention aligned to case retention
- Training on hallucination detection
- Constraints
- Cannot auto-file model text
- Local ethics rules still bind attorney of record
- Sealed/CJI material must stay off consumer models
- Security
- Treat research prompts as sensitive; encrypt logs; least-privilege access; ban training on agency case text without contract controls.
- Proof of value
- Sampled pre/post rate of unverified citations caught before filing
- Log completeness for sanctioned matters
- Shadow-IT consumer-tool usage decline
Delivery
Role takeaway
- Work
- Publish policy, configure enterprise AI, train attorneys/staff, institute sampled brief audits, and define sanctions-response playbook.
- Dependencies
- IT/DLP controls
- Clerk e-filing form changes
- Bar ethics guidance alignment
- Leadership enforcement
- Ownership
- Attorney of record owns filing accuracy; practice AI compliance lead owns audits; CIO owns endpoint controls.
- Skills and adoption
- Mandatory hallucination/cite-check training; supervisors review first 30 AI-assisted filings.
- Governance checkpoints
- Policy sign-off
- Pilot gate before office-wide enablement
- Quarterly audit report to leadership
- Acceptance criteria
- 100% of AI-assisted filings carry attestation
- Audit sample finds zero uncorrected fabricated citations in pilot window
- Documented escalation path within 24h of discovered error
- Risks
- Deadline-driven shadow use of consumer AI
- Over-reliance on sanctions as substitute for controls
- Incomplete logs when discovery or bar inquiry arrives
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Treat citation/research GenAI as unvertrusted input into CMS/e-filing pipelines. Prefer tools that force human cite-check steps and retain query/response logs for sanctions defense; do not auto-insert model text into filed pleadings.
Governance
Who approves, reviews and stays accountable for outcomes?
Courts and DA/PD offices need written GenAI use policies, mandatory verification attestations, and clear escalation when fabricated authority is discovered. Sanctions and bar referrals remain active enforcement levers.
Security and privacy
What data, permissions and controls need testing?
Consumer GenAI used for case research can exfiltrate confidential or CJI-adjacent facts; constrain to enterprise instances with logging and prohibit pasting sealed or victim-identifying material.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Training must reach solo and high-volume criminal practitioners, not only large firms; judges and clerks need shared vocabulary for spotting fabricated authority.
Procurement
What should contracts, pricing and exit terms secure?
Score vendors on verification workflow, logging, and contractual bans on training on case data—not marketing claims of legal accuracy.
Operating model
Which teams own the service once it runs?
Shift from 'AI for speed' to 'AI with mandatory verification time budgeted into filing SLAs.'
Publication history
- 2026-09-17Public Safety · Issue 114 resources
Stable resource ID: reuters-ai-court-filings-surge-20260917