Microsoft backpedals on copilot's 'entertainment only' disclaimer
Microsoft is scrambling
to revise the terms of use for its Copilot AI assistant after a viral backlash highlighted a disclaimer stating the tool was “for entertainment purposes only.” The about-face comes as the software giant aggressively pushes Copilot as a core component of its future, including ambitious forays into superintelligence, raising questions about the company’s internal confidence in its own AI.A curious caveat emerges
The jarring disclaimer, unearthed by users on X (formerly Twitter), essentially absolved Microsoft of any responsibility for Copilot’s accuracy or functionality. “Copilot is for entertainment purposes only,” the terms read. “It may make mistakes and not work as expected. Do not rely on Copilot for important advice. Use Copilot at your own risk.” This starkly contrasts with Microsoft CEO Satya Nadella's recent praise for Copilot’s “precision and low latency” during the company’s January earnings call.
The origins of the “entertainment” language trace back to February 2023, when Copilot debuted as a complementary search service within Bing. However, Microsoft now concedes that the phrasing “no longer reflects how Copilot is used today.” A spokesperson explained that the language is a legacy holdover and will be updated in the near future. The speed with which Microsoft is walking this back suggests the issue became more pressing than initially anticipated.
Beyond copilot: a broader liability landscape
It's worth noting that the Copilot terms aren't unique in their cautious wording. Microsoft’s broader Services Agreement also addresses AI without mentioning “entertainment,” and rival AI developers—OpenAI, Anthropic, xAI, and Meta—employ similar disclaimers to mitigate legal risk. OpenAI, for instance, states users must “accept and acknowledge that any use of the results of our service is at your own risk.” xAI, backed by Elon Musk’s SpaceX, goes even further, requiring users to indemnify the company against “any claim or damage,” effectively shifting almost all liability onto the user.
The current legal environment surrounding generative AI is still murky, and these disclaimers represent a preemptive defense against potential lawsuits. The emergence of cases like the dozen filed against OpenAI in California, alleging damages stemming from GPT-4o's allegedly flattering responses (including a tragic case involving suicide), underscores the escalating stakes. Nippon Insurance Company’s lawsuit against OpenAI, stemming from a client’s legal troubles after ChatGPT advised her to distrust her lawyer, further illustrates the potential for AI-driven misguidance to have real-world consequences.
But Microsoft's initial positioning, with its blunt “entertainment only” caveat, arguably reveals a deeper uncertainty. While other companies temper expectations, Microsoft seemed to actively downplay its AI's reliability, a contradiction to its public messaging.
The fine print of responsibility
Meta, in its terms, prohibits users from relying on AI results for “professional advice related to medicine, finance, law, or pharmaceuticals,” highlighting the specific areas where AI guidance is deemed unacceptable. This level of granularity demonstrates a heightened awareness of the potential pitfalls. xAI’s sweeping indemnification clause, demanding users take full responsibility for any consequences arising from its use, represents perhaps the most aggressive approach to liability avoidance.
The rush to deploy generative AI is outpacing the development of legal frameworks to govern its use. Microsoft's awkward retraction of the Copilot disclaimer serves as a potent reminder: even tech giants are grappling with the inherent risks of unleashing powerful, yet imperfect, AI upon the world. The implications of this rapid evolution are only beginning to be understood.