Amid accusations that Chinese firms like DeepSeek and Moonshot AI distilled outputs from OpenAI and Anthropic models, U.S. security agencies issued a joint advisory warning about the practice. Y Combinator CEO Garry Tan pushed back, arguing regulators should avoid restricting distillation and instead focus on balancing open-weight and frontier AI models, while suggesting smaller U.S. labs use similar techniques on domestic frontier models.
news.slashdot.org
· 2026-09-14
Researchers behind the dealignai project published a modified version of DeepSeek's V4.1-Flash model with its safety refusal circuitry surgically removed at the weight level, while preserving core capabilities like reasoning, vision, and its 1M-token context. The team says the checkpoint loads like a standard model with no special code needed, and claims HarmBench testing shows a 100% attack success rate for harmful prompts across both low and high reasoning effort settings, compared to the base model's much lower compliance rate.
huggingface.co
· 2026-09-11
Garry Tan, CEO of Y Combinator, said at the accelerator's Demo Day that regulators should take no action against AI model distillation, even as OpenAI and Anthropic accuse Chinese firms like DeepSeek, Moonshot AI, and MiniMax of copying their models' outputs to train cheaper alternatives. His comments come days after the NSA, CISA, and FBI issued a joint advisory warning about the practice.
cnbc.com
· 2026-09-11
Anthropic disclosed that Chinese AI labs including Alibaba, Moonshot and DeepSeek secretly funneled user requests through Claude and harvested its outputs to train their own competing models, a practice it calls illicit distillation. Alibaba's campaign alone involved more than 151 million exchanges between May and July, peaking near 3 million a day from over 3,500 fraudulent accounts, while Moonshot silently rerouted Kimi customer queries to Claude and presented the answers as its own.
cnbc.com
· 2026-09-11
Anthropic published findings showing five distinct campaigns, mostly linked to Chinese AI labs, that extracted nearly 200 million exchanges from Claude models to train rival systems. The largest, attributed to Alibaba, alone generated 151 million exchanges between May and recent months, using tricks like disguised translation requests to expose the model's hidden reasoning steps.
techcrunch.com
· 2026-09-10
Anthropic alleges that Chinese AI firms DeepSeek and Moonshot created thousands of fake accounts to funnel millions of real user queries into its models, a technique known as distillation used to replicate its AI's capabilities without building them independently. The company claims this activity violated its usage policies and represents a large-scale attempt to extract proprietary model behavior through automated querying.
wsj.com
· 2026-09-10
GreyNoise reports that a likely Russian-speaking threat actor deployed hundreds of AI agents using OpenAI's Codex and DeepSeek models to build and launch exploits against two PaperCut NG/MF vulnerabilities. The campaign, which began August 31, compromised at least 440 servers across 395 organizations in 48 countries, mostly in education, with credentials stolen from 280 victims and admin access gained at 12 organizations.
bleepingcomputer.com
· 2026-09-10
DeepSeek has released V4.1-Flash on its API, replacing the earlier V4-Flash and V4-Flash-Vision-Exp models. The new model adds native multimodal capabilities and can be accessed by setting the model parameter to deepseek-flash.
twitter.com
· 2026-09-10
The NSA, CISA, and FBI jointly identified six Chinese AI companies—DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI—as running large-scale campaigns since late 2024 to extract the underlying capabilities of leading US AI systems like Claude, GPT, Gemini, and Grok. The agencies said these firms likely operated with the Chinese government's awareness and used tactics such as mass fake-account API abuse and prompt injection to expose hidden reasoning processes.
arstechnica.com
· 2026-09-09
US federal agencies issued a joint advisory on Sept. 8 alleging that Chinese AI companies including Alibaba, DeepSeek, MiniMax, Moonshot AI, StepFun and Z.AI have run large-scale operations to pull outputs from US models like Claude, GPT, Gemini and Grok since late 2024. The advisory says these firms extracted billions of tokens through millions of queries, using techniques such as chain-of-thought extraction and automated evasion to dodge blocking measures, in order to train their own competing systems.
darkreading.com
· 2026-09-09
A follow-up experiment tested four open models—DeepSeek V4 Flash, Inkling, Kimi K3, and Qwen3.8 A95B—by inserting the first 1% of GPT-5.5 Pro's reasoning trace into each model's own reasoning channel before letting it generate answers freely. Researchers then measured how much of GPT-5.5 Pro's visible answer text overlapped with each model's output. Qwen3.8 showed the largest jump, with overlap rising from 33.92% unprefilled to 54.50% with the GPT-5.5 Pro prefill, a 20.58 percentage-point increase, while other models showed much smaller shifts.
gist.github.com
· 2026-09-09
CISA, the NSA, and the FBI issued a joint advisory naming DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI as having harvested billions of tokens from Anthropic, OpenAI, Google, and xAI models since late 2024. The firms allegedly used fraudulent accounts, proxy networks, and automated failover systems to bypass rate limits and extract restricted reasoning data at industrial scale.
bleepingcomputer.com
· 2026-09-09
GoKawiil