Goldman Sachs told shareholders in its annual letter: Although we use AI, generative AI models are prone to making mistakes, which can result in:

AI News


Goldman Sachs told shareholders in its annual letter: Although we use AI, generative AI models are prone to making mistakes, which can result in:
Goldman Sachs aims to deeply integrate AI into its operations and revolutionize customer onboarding, risk management, and more with new operating models. CEO David Solomon has expressed optimism about the transformative potential of AI and the bank’s strong financial performance, while bluntly warning shareholders about risks such as false outputs, data breaches and external dependencies.

Goldman Sachs wants shareholders to know two things about AI. It’s that AI is central to the company’s future, and that it can go in the wrong direction in important ways. In its 2025 annual report released Friday, the bank laid out both its AI ambitions and blunt warnings about the technology it’s betting on. CEO David Solomon struck an overall bullish tone, with net revenue up 9% year over year to $58.3 billion, EPS up 27% to $51.32, and return on equity improving 230 basis points to 15%. However, the AI ​​section of the letter was particularly appreciated.

Goldman is applying AI to six business areas it calls “ripe for disruption”

The bank announced a new operating model built around AI, so-called One Goldman Sachs 3.0. The six workstreams the company is initially targeting are client onboarding and KYC, vendor management, regulatory reporting, lending, enterprise risk management, and sales enablement. According to the letter, its ambitions go beyond just platform upgrades. It’s about completely rethinking how companies organize their workforces, make decisions, and think about productivity and resilience.Goldman also deployed the GS AI chatbot to all of its 47,000+ employees and partnered with Cognition Labs to build a custom tool.

But the bank also warns investors about what AI can get wrong.

The company’s risk disclosures are a specific part of the letter. This warns you that your generative AI model may produce incorrect output. And, in the worst case scenario, that could mean exposing private, confidential, proprietary information, or output shaped by biases built into the training data.The bank also notes that it relies on third-party AI developers, creating dependencies in how those providers build and update their models, a dependency that Goldman does not have full control over. And on the threat front, we acknowledge that bad actors could leverage the capabilities of AI to commit fraud, embezzle funds, and launch cyberattacks.The letter notes that the legal and regulatory landscape surrounding AI remains uncertain and rapidly changing.Solomon ended his talk on a still optimistic note. AI will change the way people live and work, he writes, but the speed of its adoption raises big questions and there will be winners and losers. Goldman is moving fast enough so far to want to be in the first group, but it’s being very careful to remind shareholders that it could still level out.



Source link