Researchers Extract Proprietary Reasoning Traces from Anthropic, OpenAI and Google LLMs, Revealing Hidden Secrets
A new research paper demonstrates that the hidden reasoning traces generated by leading large language models from Anthropic, OpenAI and Google can be recovered in full by using weaker models from the same providers, without attacking the target model itself.
The technique, described in the preprint "Stealing Reasoning Traces from Proprietary LLM APIs," exploits the fact that providers return encrypted reasoning blocks to clients so the model can continue multi-turn conversations. Researchers discovered that these blocks can be submitted to a less-protected smaller model, which is then prompted to reproduce the original chain-of-thought verbatim.
Testing on 120 Codeforces problems confirmed that the recovered traces matched the token counts reported in API metadata, proving the output was genuine rather than hallucinated. When the same method was applied to publicly available agent logs on GitHub and Hugging Face, analysts recovered 315,320 reasoning blocks containing 704 unique secrets, among them 62 API keys, 33 passwords and 24 access tokens that existed only inside the encrypted portions.
Multiple attack vectors identified
The paper outlines four distinct risks. First, competitors can distill high-quality reasoning from stronger models without triggering distillation defenses. Second, internal safety reasoning that is normally filtered from visible answers can be extracted, revealing detailed discussions of criminal techniques. Third, the mechanism can be reversed to embed malicious instructions inside shared logs that later get replayed by other users as part of the model’s own past reasoning.
Finally, the visible summaries returned by APIs were shown to omit or alter the actual sequence of thoughts. On certain AIME problems, the full trace revealed that the model first recalled a known answer before attempting to solve the problem, a behavior absent from the polished summary presented to users.
The authors advise developers to treat encrypted reasoning blocks as sensitive credentials and to avoid publishing or reusing uninspected agent trajectories from public repositories.
Related articles
TenetBench Benchmark Tests LLMs for Compliance with Professional Ethical Codes
Researchers from the Institute of Information Security Issues at ISP RAS and the Institute of Cybersecurity and Information Protection at RANEPA have developed TenetBench, a benchmark designed to evaluate how well large language models adhere to specific norms in professional codes of ethics. The benchmark generates realistic work-related prompts that may conflict with rules from documents such as the Code of Professional Ethics for Advocates or journalistic standards on undisclosed advertising. Each test case pairs a conflicting request with a neutral counterpart to measure whether models refuse, warn, or still provide prohibited content. Evaluation separates task completion, norm violation detection, and presence of explicit refusals using an LLM-as-a-Judge approach validated against human annotators. The full dataset contains 3,433 prompt pairs derived from 15 professional codes, with a public subset of 2,235 pairs from 10 documents. Human validation confirmed 99.5% of generated examples as fully plausible work scenarios.
Solar to Showcase Solar AI Firewall for Business Protection Against AI Risks at SOC Forum 2026
GK Solar will present its new Solar AI Firewall solution at SOC Forum 2026 to help organizations control interactions with AI services and prevent confidential data leaks through employee prompts. The product addresses growing risks from staff sending corporate data to neural networks and AI agents gaining access to internal systems. The forum takes place on 27-28 October in Moscow's Lomonosov cluster, with the official launch of Solar AI Firewall scheduled during the event. Solar will also highlight practical risks of AI adoption, including data transmission to external services, use of public models by employees, and new attack surfaces in corporate infrastructure. The solution will be demonstrated as part of an integrated ecosystem for access, network, and data protection. In addition, the company plans to introduce JSOC ONE, a new line of Solar JSOC services tailored for mid-sized businesses that require 24/7 monitoring without building their own SOC.
Researcher Runs 100 Abliterated AI Agents Against His Own Accounts, Achieves Five Successful Breaches
A security researcher deployed roughly 100 autonomous AI agents based on abliterated open-source models to attempt unauthorized access to his online accounts over a five-hour period. The agents successfully compromised three accounts by exploiting vulnerabilities in legacy personal projects and two additional accounts through password reuse and credential stuffing. They also conducted 16 social engineering attempts, harvested and cross-verified personal data from public sources, and generated phishing pages targeting the researcher's Substack and Hacker News presence. Despite these successes, the agents failed to discover any zero-day vulnerabilities in third-party software or gain access to Tier-0 accounts such as Gmail, 1Password, or banking services. The experiment relied on abliterated versions of GLM-5.3 and DeepSeek V4 Flash models running on cloud GPUs, orchestrated via Codex CLI inside Docker containers with access to Railway infrastructure. Total compute cost reached approximately $210, equating to roughly $40 per successful account compromise. The researcher concluded that inexpensive swarms of such agents could soon enable low-skill attackers to systematically map and exploit an individual's entire online footprint.
How Censorship in Closed AI Models Reshapes Science: Hugging Face Releases Open Carbon-X DNA Model
Hugging Face CEO Clement Delangue announced the release of Carbon-X, an open 14.8 billion parameter language model designed for reading, analyzing, and generating DNA sequences. The announcement highlighted how shallow guardrails in closed commercial models blocked even proofreading a tweet discussing the topic. Commenters and researchers pointed out that these input/output filters lack context awareness, blocking legitimate scientific queries while motivated adversaries bypass them through rephrasing. Carbon-X was built on Qwen3-14B with a hybrid BPE and 6-mer tokenizer, domain mid-training on 302 billion tokens, and safety achieved through data curation rather than runtime censorship. The model achieves performance comparable to closed frontier systems at roughly 1/1800th the inference cost for large-scale genomic tasks. The case underscores growing asymmetry between closed corporate AI labs building wet labs and independent researchers restricted by superficial safety layers.