All news with #model theft tag

🔒 OpenAI disclosed it disrupted a coordinated distillation campaign that illicitly attempted to extract protected reasoning from its models. The activity, traced to early July 2026 and linked by OpenAI to individuals associated with Moonshot AI, scaled in late July before being halted on July 28. OpenAI described the attack as adversarial distillation and implemented mitigations, closed a replay pathway, and banned fraudulent accounts to prevent further extraction and replay of encrypted reasoning.

🔒 Security researchers report a surge in proxy "transfer stations" used to hide origin of requests to frontier AI models, enabling credential abuse and model distillation. Team Cymru identified tens of thousands of relay servers running open-source platforms like CRS and sub2api that authenticate users with pooled stolen API keys and subscriptions. Activity links include China and Hong Kong, with many relays hosted on U.S. VPS providers, and evidence suggests heavy upload traffic consistent with extraction attempts. Organizations are advised to treat AI credentials as sensitive secrets, use short-lived tokens, monitor usage, and have playbooks to revoke compromised keys.

🔐 Google’s GTIG reports that state-affiliated groups and cybercriminals are increasingly targeting AI-related assets — from model weights and proprietary source code to API keys and cloud compute — to support espionage, extortion, and resource theft. The quarter’s incidents included exfiltration of proprietary models, distillation campaigns using hundreds of millions of prompts, and cloud compromises that enabled attackers to run unauthorized AI workloads. Attackers also deploy automated, agent-driven frameworks to scale reconnaissance, credential harvesting, and exploitation.

🛡️ Anthropic says it identified and disrupted industrial-scale illicit distillation campaigns run by seven China-based labs, including Alibaba, Moonshot, DeepSeek, Z.ai (Zhipu), and MiniMax. The company reports attackers used proxy networks, fake accounts, stolen payment credentials, and harvested API keys to stealthily route user queries through Claude and save transcripts for training. Anthropic observed large-scale extraction of chain-of-thought, agentic capabilities, coding, and reasoning data and has updated Claude and its policies to limit such misuse.

🛡️ Two recent episodes involving Anthropic’s Claude — a China-based large-scale extraction campaign and the U.S. government’s ban of Claude for federal use — expose a growing operational risk in frontier AI. When adversaries can probe a model at scale they can map strengths and predictable seams, and when major customers demand behavioral changes vendors face immediate, operational trade-offs. CISOs should treat frontier models as high-value dependencies shaped by upstream pressures and invest in visibility, monitoring, and governance to detect when external influences begin affecting deployed behavior.

🔒Anthropic said three China-based AI firms — DeepSeek, Moonshot and MiniMax — executed more than 16 million exchanges with its Claude model using roughly 24,000 fraudulent accounts to perform model distillation, breaching terms of service and regional access restrictions. The company described the activity as deliberate capability extraction and attributed the campaigns via IP address correlation, request metadata and infrastructure signals. To mitigate further misuse, Anthropic has implemented detection systems for API attack patterns, tools to detect chain-of-thought elicitation and coordinated account activity, stronger verification for high-risk accounts and product-, API- and model-level safeguards.

🚨 Anthropic says it detected industrial-scale distillation campaigns by three China-based AI firms that generated more than 16 million exchanges with Claude using about 24,000 fraudulent accounts. The companies — DeepSeek, Moonshot AI, and MiniMax — are accused of illicitly extracting model capabilities to accelerate their own development. Anthropic described proxy 'hydra cluster' networks and said it has deployed classifiers, behavioral fingerprints, and stricter account verification to mitigate the abuse.

🔐 Google’s newest GTIG AI Threat Tracker outlines rising adversarial misuse of AI, documenting how threat actors are distilling models, experimenting with agentic capabilities, and integrating AI into malware and social engineering. The report highlights activity from groups including APT31, North Korean and Iranian actors, and malware families such as HONESTCUE. It underscores growing risks from model extraction, the emergence of illicit jailbreak services like Xanthorox, and recommends that AI providers monitor API access and adopt robust defenses.

🔒 Google detected and blocked a coordinated campaign of more than 100,000 prompts it says were designed to extract the proprietary reasoning capabilities of its Gemini model. The requests resembled model-extraction or distillation efforts intended to reproduce internal reasoning traces across multiple languages, and Google says it intervened in real time to reduce the immediate risk. The company cautioned that systematic extraction can amount to intellectual property theft and said it will pursue takedowns and legal remedies while trying to balance legitimate research and evaluation needs.

🛡️ Google Threat Intelligence Group (GTIG) reports rising adversarial use of AI in Q4 2025, including widespread model extraction, AI-augmented reconnaissance, social engineering, and trials of agentic tooling. GTIG and Google DeepMind detected and mitigated numerous extraction attempts, protected internal reasoning traces, and disabled abusive assets in real time. The update describes AI-enabled proofs-of-concept (for example HONESTCUE and COINBAIT), abuses of shareable chat outputs, underground proxy toolkits, and published IOCs to support defenders.

🛡️ Linwei Ding, a former Google engineer, was convicted by a federal jury on multiple counts of economic espionage and theft of trade secrets after allegedly taking more than 2,000 confidential documents tied to Google's AI infrastructure and chip designs. Prosecutors say the material included details on Google's TPU and GPU architectures, Cluster Management System software, and custom SmartNICs used in AI supercomputers. Authorities allege the theft occurred between May 2022 and April 2023 and that Ding copied files to personal accounts and founded a China-based startup while still employed by Google. He faces significant federal prison terms if sentenced.

🔒 AI systems face a growing range of attacks—from data poisoning and model poisoning during training to adversarial inputs, prompt injection, and model theft during deployment. These threats exploit weak data governance, supply chain dependencies, and inadequate monitoring. Security leaders should adopt proactive controls including provenance tracking, adversarial testing, rate limits, and routine red teaming. Frameworks like MITRE ATLAS can help map attacker techniques and prioritize defenses.