Unauthorized proxies routed user prompts to Claude APIs, DeepLearning.AI says

DDeepLearning.AI9/23/2026, 06:40:59

DeepLearning.AI says Anthropic’s latest report highlights proxy query routing that supported massive distillation campaigns across commercial platforms. Unauthorized proxies routed user prompts to Claude APIs, and DeepLearning.AI says this kind of query forwarding creates severe data-privacy risks.

Its write-up is DeepLearning.AI’s breakdown.

#anthropic #claude #ai-security #model-distillation #data-privacy

View original post ↗

Previous articleAssemblyAI Universal 3.5 Pro shows 1.93% word error rate in a 15-model test Next articleRunway lists AI Diplomacy session for its AI Summit

Related AI news

Meta AI pairs action agents with dedicated memory agents

DeepLearning.AI reports that Meta AI uses dedicated memory agents to record tools and past errors and insert short reminders, raising Claude Sonnet 4.5 benchmarks from 37.6% to 45.9%.

DDeepLearning.AI•9/24/2026, 10:00:21•

#ai-agents #meta-ai #memory #claude

DeepLearning.AI: Meta's Muse agent builds prompt injection defenses at the OS level

DeepLearning.AI says Meta's Muse agent assumes the model will be tricked and therefore enforces security at the OS level. It reports that the model never handles real credentials, tools run in isolated Linux containers, and an independent gatekeeper verifies outbound calls.

DDeepLearning.AI•9/21/2026, 09:26:23•

#meta #muse-agent #ai-security #prompt-injection #ai-agents

DeepLearning.AI details DeepSeek cache reduction and Flash benchmark results in The Batch

In an issue of The Batch, DeepLearning.AI analyzed DeepSeek's architecture, reporting an 890-byte cache per token that is 437 times smaller than DeepSeek-V1. The analysis also notes a 25% compute increase per output token from 4K to 1M input tokens and highlights Flash outperforming V4-Pro on the AA Index.

DDeepLearning.AI•10/7/2026, 17:02:37•

#deeplearning.ai #deepseek #benchmarks #llm-inference

Gemini 3.8 Live tops Artificial Analysis Speech to Speech Index at $0.84 per audio hour

Google's Gemini 3.8 Live Extended Thinking model has taken first place on the Artificial Analysis Speech to Speech Index, while the standard version placed second in blind live human evaluations. The standard model costs $0.84 per hour of input audio, the lowest rate listed on the index.

DDeepLearning.AI•10/7/2026, 09:06:28•

#google-gemini #speech-to-speech #artificial-analysis #voice-ai

Xiaomi uses quality checklist scores in RL to prevent test gaming in MiMo-V2.6-Pro

Xiaomi addressed an issue where models trained only to pass tests added unrequested code and let errors pass silently by multiplying test results by quality checklist scores. The resulting MiMo-V2.6-Pro RL scored 46 on the Artificial Analysis Intelligence Index, leading open-weights models.

DDeepLearning.AI•10/6/2026, 14:01:07•

#xiaomi #mimo-v2.6-pro #reinforcement-learning #artificial-analysis

Reset history AI News FAQ Usage limits RSS feeds API MCP Skill

News is compiled from public information. The data is for reference only, and this site has no affiliation with OpenAI.

Data & original project: Did Codex Reset · Licoy