API security flaw highlights AI model vulnerabilities
Researchers uncover a way to make weaker AI models reveal the hidden reasoning of more powerful systems.
A new paper by a team of researchers has revealed critical vulnerabilities in how major large language model (LLM) providers encrypt and manage hidden reasoning traces within their proprietary APIs.
The paper, titled “Stealing Reasoning Traces from Proprietary LLM APIs,” details how bad actors can access and steal proprietary “chain of thought” reasoning from Claude, ChatGPT, and Gemini models—the step-by-step problem-solving material (or reasoning traces) that parent companies Anthropic, OpenAI
Only users who have a paid subscription or are part of a corporate subscription are able to print or copy content.
To access these options, along with all other subscription benefits, please contact info@waterstechnology.com or view our subscription options here: https://subscriptions.waterstechnology.com/subscribe
You are currently unable to print this content. Please contact info@waterstechnology.com to find out more.
You are currently unable to copy this content. Please contact info@waterstechnology.com to find out more.
Copyright Infopro Digital Limited. All rights reserved.
As outlined in our terms and conditions, https://www.infopro-digital.com/terms-and-conditions/subscriptions/ (point 2.4), printing is limited to a single copy.
If you would like to purchase additional rights please email info@waterstechnology.com
Copyright Infopro Digital Limited. All rights reserved.
You may share this content using our article tools. As outlined in our terms and conditions, https://www.infopro-digital.com/terms-and-conditions/subscriptions/ (clause 2.4), an Authorised User may only make one copy of the materials for their own personal use. You must also comply with the restrictions in clause 2.5.
If you would like to purchase additional rights please email info@waterstechnology.com
More on Emerging Technologies
On vibe coding, no/low-code dev, and some potential misconceptions
The Waters Wrap: Is no-code/low-code even a thing anymore? Not really. But the companies that championed those terms just a few years ago tell Anthony they aren’t going anywhere.
Repo tokens won’t be cleared. Or will they?
Uncertainty lingers over clearing status of tokenized Treasuries, with decision likely devolved to DTCC.
AI startups get big-name investors, SEC denies 24X’s exemption request, and more
The Waters Cooler: A recap of the major tech and data news from the past week in the capital markets.
Rapid-fire repo raises hopes of cheaper, faster trading
Tokenized Treasuries piloted by DTCC could squash settlement cycles and enable 24/7 repo.
Stopgaps and fail-safes: How trading vendors guard against rogue AI agents
TradeStation’s newest trading tool, Titan-X, uses agentic tools to help plan, execute, and collect data on trades.
A rookie’s guide to tokenized Treasuries
What are DTCC’s digital US government debt securities? How do tokenized repo and other transactions work? These questions, and others, answered.
The complexity of using AI to tackle compliance
The Waters Wrap: Law firms are introducing new tools to help with regulatory compliance, potentially encroaching on regtech vendors’ territory, Wei-Shen writes.
Anna’s new digital token identifiers, TS Imagine offers prediction market data, and more
The Waters Cooler: A recap of the major tech and data news from the past week in the capital markets.
