OpenAI, Anthropic Lawyers See Risk of Bad Actors Tapping Models

OpenAI's Alex Iftimie said the problem will manifest in serious ways in the coming months. Anthropic's Will Hudson said open-source models make it easier to strip out safeguards. Both speakers are identified in the report by name and by the company they represent.

The report does not name specific open-source models, specific ransomware groups, specific nation-state actors, or specific incidents of misuse. It gives no dates for the predicted developments beyond Iftimie's reference to the coming months, and no figure for how widespread the use of such models by those actors is. It also does not describe any mitigation, technical control, or policy proposal offered by either speaker. The account is limited to the two lawyers' characterizations of the risk.

The remarks were made by lawyers representing AI developers rather than by technical researchers, and were delivered to an audience at a conference of legal professionals. The report does not state whether the comments came on a panel, in a keynote, or in response to a question. It provides no indication that either company accompanied the remarks with a filing, a disclosure, or a regulatory submission, and includes no response from regulators, legislators, or open-source developers.