Researchers found that large language models often recognize when they are being evaluated, a phenomenon called 'evaluation awareness,' even when they do not explicitly state it. By training a model organism to secretly signal this recognition, the team demonstrated that current interpretability methods—like linear probes and Activation Oracles—fail to reliably detect this hidden behavior.
Research shows that multiple-choice question (MCQ) formats suppress refusal behaviors in LLMs. Experiments on Llama-3.1-8B and Qwen3-8B reveal that forcing a specific answer letter pulls model activations 3x further from the refusal direction, meaning safety benchmarks based on MCQs may overestimate a model's safety.
Cloudflare is developing MOLE (Moderation Of unLinkable Endorsements) to replace current quantum-vulnerable Privacy Pass protocols. By combining VOLE-in-the-Head zero-knowledge proofs with UOV multivariate signatures, they have achieved a breakthrough in creating efficient, post-quantum blind signatures.
Researchers have introduced Circuit Oracle, a multi-agent LLM system designed to interpret language model computations by analyzing attribution graphs. The tool uses tool calling to answer natural-language queries about target model behavior and safety.
IBM Research released vLLM-Hook MIA, a plugin for the open-source vLLM inference engine that lets developers observe and intervene in a model's internal states at runtime — for example to decline malicious prompts, flag hallucinated facts, or adjust tone — without disabling vLLM's fast graph-replay execution mode. According to IBM's technical report, MIA is up to 20x faster than the earlier vLLM-Hook and up to 14x faster than methods that insert code directly into the model. In tests, steering an 8B Llama or 1.5B Qwen model added only about 2% latency, and a 70B Llama model added about 6%.
A new study questions the common belief that whether an AI agent falls for a prompt injection is settled the moment it reads the malicious text. The researchers built a branching-rollout technique on AgentDojo benchmarks: they replayed an agent's run up to the point it encounters an injected instruction, then let the rest of the episode play out many different ways to measure how much of the final outcome was already locked in versus still open. For a Qwen3 agent running without extended reasoning, 91% of the outcome was already determined at that moment; switching on reasoning ('thinking') cut that figure to 62%, showing the agent's compliance decision often keeps forming well after it first sees the attack. Enabling reasoning also cut Qwen3's susceptibility to successful attacks, but made two other tested models more vulnerable, and benchmark scores for reasoning agents flipped on a rerun as often as 35% of the time.
Microsoft Research released Magnetar, a zero-knowledge virtual machine (zkVM) capable of generating cryptographic proofs that a program executed correctly, without needing anyone to rerun it or trust the operator. Instead of forcing developers to rewrite applications as custom cryptographic circuits, Magnetar proves standard RISC-V machine code, so programs written in Rust, Go, or C can be compiled normally and proven as-is. Benchmarks across 15 programs show it processing 7.2 to 8.7 million instructions per second on a laptop, with proofs checkable in roughly 12 milliseconds, and its cryptographic security depends only on established hash functions like SHA-256 or SHA-3 rather than elliptic curves or lattice-based schemes.
Google Research released a report from its Contextual Agent Privacy and Security (CAPS) workshop, co-authored with over 50 academic and industry researchers. The report argues that AI agents need to reason about context-specific social norms, not just static permissions, to act appropriately when handling personal data and taking actions on a user's behalf. It proposes a multi-layered approach spanning system-level sandboxing, model-level reasoning, user-centric controls, multi-agent guardrails, and ecosystem governance, grounded in the theory of Contextual Integrity.