Archive
模型安全、資安、隱私、供應鏈、紅隊與防護。
代理工具
StateBridge uses an orthogonal Procrustes transformation to map the sending agent’s final-layer hidden states into the receiving agent’s input embedding space, without training a projector. Its average scores across four models exceeded the strongest baseline by 2.4 to 2.9 points, but it currently supports only agents using identical model weights.
最新模型
AlayaWorld has rewritten the spatial memory and conditioning pipeline of its interactive world model so that camera controls, historical frames, and generated content use aligned causal VAE representations. A new report claims the highest overall consistency score on WBench, but the updated weights and inference code have yet to be released.
AI 安全
Using independent manual and agent-based tracing, VICBench labels the vulnerability-introducing commits for 100 CVEs across Python, Java, and C++ projects. Cross-file moves and refactoring cause extensive misclassification by V-SZZ and LLM4SZZ, while also exposing reliability problems in affected-version data.
AI 安全
Anthropic will add model-level text watermarks to Claude models released after August 2, 2026, while embedding signed C2PA provenance data in supported files. The watermark can only indicate that content may have been processed by Claude; it cannot prove that a work was originally created by AI. No public detection tool or empirical accuracy results have yet been released.
AI 安全
IO Factory connects narrative planning, agent posting, platform exposure, audience-state updates, and baseline comparisons into a traceable simulation lifecycle. The system has completed runs with 100,000 simulated civilians and 10,000 operator agents, but all influence measurements derive from manually configured rules and LLM judges.
最新模型
A Heidelberg University study finds that although individual latents in sparse autoencoders may be interpretable, the full set of activated latents does not map reliably onto human concepts. Simply adding a compatible adjective before a noun can make 20% to 60% of the original activation set disappear, limiting the reliability of interpreting model states through set overlap.
最新模型
Meta has released the Muse Glimmer 30B weights, vision encoder, and DFlash draft model. The 17GB version can run text, image, and speculative decoding workloads simultaneously within 24GB of VRAM. Meta measured an average of 233.4 tokens/s on an RTX 5090, but safety evaluations and deployment details show that the model still requires external safeguards and a newer runtime.
最新模型
XPeng introduces XCoT-VLA, replacing dozens of natural-language reasoning tokens with a limited set of semantic action tokens, then using flow matching to generate continuous trajectories. Open-loop tests show lower lateral error, but no closed-loop safety evaluation has been reported, and the latency figures do not cover the full in-vehicle hardware and software stack.
AI 安全
REDAgentBench executed 1,661 attack cases across sandboxed workspace, email, browser, banking, and external file services, rather than judging harm solely from an agent’s response. When the same runs were scored using service receipts and final states, the attack success rates of all six models were 7.73 to 11.72 percentage points higher than with trajectory-only scoring.
AI 安全
A preregistered reproducibility audit found that a vulnerability PoC that executes and triggers a crash does not necessarily reproduce the specified CVE. In 20 of the 30 evaluable cases, the same signal still appeared in the patched version, showing that agentic security benchmarks need stricter oracles than strings and exit codes.
最新模型
Universal Activation Bus aligns the intermediate representations of five language models into a fixed 3,072-dimensional space, requiring new models to train only their own linear encoder and decoder. In experiments, models could reuse the same probes, sparse autoencoder, and natural-language explainer, although representation reconstruction for wider models remained noticeably distorted.
AI 安全
ElasticBack hides malicious rules in `SKILL.md` and uses specific concepts in natural-language prompts as activation switches, without modifying model weights. Experiments show that the attack can evade multiple static and runtime checks, although false-trigger rates on Claude still reach 20% to 24%.