BECAUSE OF AI, home

Measured AI benefits · as of

87% vs 27%of grasp trials kept within the force limit with the AI controller on versus off, one participant
receipt

Because of a deep reinforcement-learning grip controller in a brain-computer interface, a man with complete tetraplegia kept his grasp force within the set limit in 87% of trials, against 27% without it.

receipt

In a first-in-human trial at the Feinstein Institutes, a man with complete tetraplegia used a brain implant whose decoder and deep reinforcement-learning agent drove stimulation of his own hand, and grip control was compared with the agent on and off.

Sources

Editor's notes

verify: the text gives no calendar dates for the experiments (only the 2020 injury and a trial spanning more than 3 years), so achieved dates and window are left empty. verify: the published date is the press date of 16 July 2026; the text read shows acceptance on 2 June 2026 and no day of publication. verify: the trial is registered (NCT03680872) but this is a single participant, so the evidence grade is D.

History

  • Added by the backfill run.

Verification

Drafted by
Anthropic claude-sonnet-5-5
Verified by
OpenAI gpt-6.1-sol · pass
How sure
Speculative
  • Not dated yet (The calendar timing of the measured grasping benefit cannot be established from the supplied text.).
  • The two checkers disagreed on attribution: the drafter said isolated and the verifier said ai-vs-ai. The entry shows the drafter's grade, isolated; treat that grade as unsettled.
Record
data/verifications/feinstein-double-neural-bypass-rl-grasp-tetraplegia/2026-11-18.json · The paper supports 87% versus 27% mean session-level grasp success in one participant with RL on versus off. Both conditions used an AI decoder, and the supplied text does not establish when these experiments occurred.

Two models from different vendors read the primary source independently; this entry is published because they agreed on every receipt and every grade. It is marked speculative: The source is a press release, a news report, or the organisation's own announcement, or the benefit has not yet been measured against anything or dated, or the two checking models disagreed on a grade. Every number is still quoted from the source, and both checking models support it. The models grading an AI registry have an interest in the answer. Cross-vendor checking and the receipts on this page are how the registry handles that, not a claim to have removed it. How the gate works · Dispute this entry

Cite this entry

Permalink https://becauseofai.org/entry/feinstein-double-neural-bypass-rl-grasp-tetraplegia“Because of a deep reinforcement-learning grip controller in a brain-computer interface, a man with complete tetraplegia kept his grasp force within the set limit in 87% of trials, against 27% without it.” Because of AI, updated Nov 18, 2026. https://becauseofai.org/entry/feinstein-double-neural-bypass-rl-grasp-tetraplegia