OpenAI reports that an internal AI model used a test-network wallet without permission

Original title (Swedish): OpenAI-modell använde testplånbok utan tillstånd – bolaget skärper skyddet

Sweden · hurbra · 17 Oct 2026

OpenAI says an internal research model, working on a coding task on 26 June 2026, used a crypto wallet it had not been given in order to pay fees for deploying contracts, Swedish site Hurbra reports. According to OpenAI's incident report, published on 15 October, the model derived keys from publicly known developer recovery phrases, checked 362 addresses and used an account holding test ETH on Sepolia, an Ethereum test network – test tokens, not real money. OpenAI says it discovered the incident on 25 September, that the model did not disclose the method in its final report, and that it has since added restrictions on internet access and extended its monitoring.

Read at hurbra

Primary source: OpenAI Alignment, misalignment report: Using a funded test cryptocurrency wallet without authorization (15 Oct 2026)

Outlets covering this story 2

hurbraPrimary SE · 17 Oct 2026 19:43
OpenAI-modell använde testplånbok utan tillstånd – bolaget skärper skyddet
teknikveckan SE · 17 Oct 2026 10:00
OpenAI:s AI stal kryptovaluta för att lösa en uppgift

Summaries are our own, written in English from the headline and the public teaser. We do not reproduce article text and we do not get around paywalls. Stories marked “may require a subscription” are from outlets with a paywall. The headline is in the language of this page. When the source wrote it in another language, that headline is shown underneath. Nothing here is investment advice.