Crypto
Home›Crypto›Market Structure›AI advisers can be nudged into boosting Bitcoin via hi…
AI advisers can be nudged into boosting Bitcoin via hidden prompt effects
In a June preprint, researchers said altering an internal model feature shifted the recommended Bitcoin allocation by 5.2 percentage points when amplified, and by -4.6 points when suppressed.
An AI financial adviser can recommend very different Bitcoin weights for the same client, even when the client’s finances and risk tolerance remain unchanged, according to findings highlighted by CryptoSlate.
The study had researchers audit eight frontier models and then run deeper tests using an open Google model, which they said contains an internal feature that responds selectively to Bitcoin-related concepts.
When that feature was amplified, the model’s suggested Bitcoin allocation rose by 5.2 percentage points, and when it was suppressed, the allocation fell by 4.6 points, with the intervention carried out through internal activity rather than changes to the prompt.
CryptoSlate said the paper, based on a June preprint by Wenbin Wu and co-authors, is not yet peer reviewed and the results are shown within a defined experimental setup, but the authors argue the work points to how fast personalized advice can become difficult for banks and investment firms to audit when model associations enter the allocation process.
Latest closeBitcoin $65,169.12 ▲0.4%