# PIR-2026-0001 - Benavides v. Tesla: $243M verdict over fatal Autopilot crash, upheld post-trial

- `id`: PIR-2026-0001
- `title`: Federal jury holds Tesla 33% liable for a fatal 2019 Autopilot crash; $242.57M against Tesla survives post-trial motions
- `date_occurred`: 2019-04-25 (crash, Key Largo FL)
- `date_detected`: immediate (fatal collision)
- `date_disclosed`: jury verdict 2025-08-01 (S.D. Fla., Miami); verdict upheld 2026-02-20 (Judge Beth Bloom); appellate review continuing
- `status`: corroborated (verdict and post-trial ruling on public record; dollar amount not final pending appeal)

### The agent
- `agent_description`: Tesla Model S with Autopilot engaged - Level-2 driver-assist (lane keeping + traffic-aware cruise) marketed under branding the jury found overstated its capability; human driver remains the formal operator.
- `operator_type`: individual (driver) atop an enterprise-provided automation stack
- `autonomy_level`: human-on-the-loop (L2: automation steers/accelerates continuously, human nominally supervising)
- `model_stack`: Tesla Autopilot (2019-era hardware/software; exact version not public)
- `harness`: Tesla vehicle firmware
- `failure_locus`: agent-reasoning (primary - the design failures the jury priced: no enforcement of the operational design domain, no stop-sign response; contributing operator locus: distracted driver carried 67% of fault. Previously recorded as "shared" - the fault split lives here in prose, one primary token per v0.2)

### Authority
- `authority_scope`: physical systems (vehicle control at speed on public roads)
- `funds_at_risk_usd`: unknown (bodily-injury liability; effectively unbounded per-incident)
- `blast_radius`: public (third parties outside the vehicle were the ones killed and injured)

### The failure
- `root_cause`: plain-error (primary - no adversary; the automation continued at speed through an intersection it was not designed for); contributing operator-error (driver inattention; use outside the operational design domain that the design permitted)
- `mechanism`: Driver George McGee, with Autopilot engaged on a road outside its design domain, dropped his phone and looked away. The car went through the intersection at ~62 mph and struck a parked Chevrolet Tahoe, killing 22-year-old Naibel Benavides Leon and severely injuring Dillon Angulo. The Miami federal jury (2025-08-01) allocated 33% of fault to Tesla, resting on Tesla permitting Autopilot use outside its operational design domain and overstating its capability. Award against Tesla: $242.57M - $19.47M compensatory to the Benavides estate + $23.1M compensatory to Angulo (Tesla's 33% share of ~$129M total compensatory) + $200M punitive. On 2026-02-20 Judge Beth Bloom denied Tesla's post-trial motions, ruling the evidence "more than supports" the verdict.
- `adversary_present`: no
- `exploitation_status`: in-wild-malfunction (real crash, real deaths, no attacker - reclassified from in-wild-exploited, which implies an adversary; adversary_present is no)

### Impact
- `severity`: catastrophic (one death, one severe permanent injury)
- `direct_loss_usd`: 242,570,000 judgment against Tesla - not final; appeal continuing as of Aug 2026
- `indirect_loss_usd`: unknown (litigation exposure across the wider Autopilot docket)
- `downtime`: n/a
- `data_exposure`: none (vehicle telemetry recovery was litigated at trial)

### Detection and recovery
- `detected_by`: third-party (crash scene; liability established only through six years of litigation)
- `time_to_detect`: immediate
- `time_to_recover`: n/a (fatal outcome; legal resolution still running seven years on)
- `remediation`: none specific to this crash disclosed by Tesla
- `structural_fix`: none on public record tied to this case (geofencing/ODD enforcement remained the central unadopted fix argued at trial)
- `controls_that_worked`: none identified - neither driver-monitoring nor ODD restriction prevented or mitigated the crash

### Legal
- `liability_holder`: split by jury - Tesla 33% (design/marketing), driver 67%; punitive award falls on Tesla alone
- `precedent_set`: first major federal jury verdict against Tesla over Autopilot to survive post-trial motions; establishes that fault can be allocated to the automation's design even with a distracted human formally in control
- `sealed_material`: no (trial record public)

### Evidence
- `telemetry_grade`: operator-logs (vehicle/fleet telemetry in Tesla's custody; availability and completeness of crash data were themselves contested at trial)
- `sources`:
  - https://www.cnbc.com/2026/02/20/tesla-loses-bid-toss-243-million-verdict-fatal-autopilot-crash-suit.html
  - https://www.usnews.com/news/top-news/articles/2026-02-20/us-judge-upholds-243-million-verdict-against-tesla-over-fatal-autopilot-crash
  - https://www.carcomplaints.com/news/2026/benavides-v-tesla-judge-upholds-243-million-jury-verdict.shtml
  - https://electrek.co/2026/02/20/tesla-has-to-pay-historical-243-million-judgement-over-autopilot-crash-judge-says/
  - `independence`: multiple independent outlets reporting a public court ruling - strong. (All four URLs verified resolving, v0.2 pass 2026-08-15.)
- `confidence`: high on verdict, breakdown, and post-trial ruling; the loss figure is conservative-flagged as non-final (appellate review pending).

### Verification notes
All intake facts verified, with the damages breakdown sharpened: $242.57M = $19.47M (estate) + $23.1M (Angulo) compensatory + $200M punitive, matching the intake's "~$43M of ~$129M compensatory + $200M punitive." Boundary aspect (L2 driver assist, human formally in control) retained because the verdict itself maps liability onto the automation's design - the legal block is the point of the record.