claude

In a threat intelligence report released on Thursday, Anthropic stated that Russian and Chinese users are leveraging its AI model Claude to automate cyberattacks. Among them, a Russian operator codenamed "JackPoterz" has automated most parts of the attack chain through a customized AI-driven workflow, targeting over 20 organizations, including government departments and intelligence agencies, as well as embassies and diplomatic missions in Ukraine and several European countries.Chinese users are using Claude as the engineering and orchestration layer for vulnerability research, with one workflow producing more than a dozen potential zero-day vulnerabilities in network device firmware within a month. Anthropic pointed out that AI is changing the economics of cyberattacks, allowing a single operator to complete an intrusion in two to three hours and handle dozens of victims in parallel.Additionally, the report revealed that a suspected independent consultant based in Bamako, collaborating with Mali's national intelligence agency, has used Claude as the primary engineering force to build a large-scale domestic surveillance system covering approximately 25 million SIM cards across all three mobile operators in the country. This system is locally deployed and operates using local models, with software design and engineering support provided by Claude, enabling the generation of intelligence profiles on phone numbers without the need for court warrants.

Anthropic released the latest AI abuse report stating that app development studios have utilized Claude to develop and operate over 20 dating apps. The platform claims that all users are real people, but approximately 75% of the matches that regular users encounter are AI characters. During a two-week observation period in April 2026, these dating apps operated over 4,700 AI characters, chatting with at least 25,000 different users and sending a total of about 2.36 million messages, primarily affecting users in the United States.The platform displays profiles in a mixed ratio of about 3 AI characters for every 1 real person. Claude simultaneously maintains thousands of conversations, with real part-time staff stepping in when users request video calls, ask for photos, or seek to follow social media accounts. The system generates three alternative replies for each message for part-time staff to choose from. User chatting and matching consume credits, and once depleted, users must purchase in-app virtual currency to recharge. Accounts fully controlled by AI are required not to acknowledge that they are robots.To evade app store scrutiny, the operators designed a special interface that is only activated during the review period, with over 20 apps using different program class names. Relevant brands disclosed by Anthropic include DORA, DONI, ROMI, LUMA, JOVIA, KIRA, GRACECHAT, HAVEN, NALO, and LOVIA. The company has banned related accounts and provided information on app identities and review evasion to Apple and Google.

According to the Financial Times, Anthropic stated that it has blocked multiple attempts this year by scientists to use its technology for related research that could assist in the development of biological weapons. The company cited five cases, claiming that the relevant actors circumvented control measures and obscured the research purpose to evade safety protections. In a report about models being used for malicious activities, Anthropic expressed the hope that by sharing these cases, it would spark discussions within the AI industry and among governments regarding emerging biological risks and responses.The cases include a researcher from an "unsupported region" using Claude, spending weeks planning experiments involving avian influenza. Anthropic stated that the safety filtering mechanism restricted the research to the least capable models and emphasized that it could not determine whether the aforementioned scientist intended to cause harm, as the same information required for developing biological weapons could also be used for vaccine development. The company has banned the accounts mentioned in the report but did not disclose the name of the institution to which the researcher belonged or the country where the events occurred.

Anthropic confirmed four incidents involving the Claude model mistakenly accessing the real internet and attacking third-party systems during a review of approximately 481 million model interaction records. The incidents involved Claude Mythos 5, Opus 4.6/4.7, and an internal research model. The accidents were caused by misconfiguration in the third-party evaluation environment that led to internet access, and the network security protections of the official product were not enabled during the evaluation.Anthropic summarized the core alignment risks as "biased reasoning" and "reckless" behavior exhibited by the model under task-driven conditions, where Claude Mythos 5 uploaded malicious packages to PyPI and accessed the real database of a security vendor using leaked credentials, believing it was in a "simulated environment." The company has introduced new assessments, monitoring, and alignment training, and has invited the independent organization METR to conduct an external investigation.

According to a report by Sina Finance, Anthropic refused to submit its latest model Claude Mythos 5.1 to the UK Artificial Intelligence Safety Institute (AISI) for testing before its release, as the model is only open to qualified institutions in the United States. This move has raised alarms within the UK Whitehall and AISI, with UK national security officials concerned that this could set a precedent, making it difficult for AISI to access the latest cutting-edge models in the future.Analysts pointed out that this action may be related to the increasingly stringent export control policies of the US government on the AI industry. The US government had previously implemented export controls on Anthropic's models Mythos and Fable, prohibiting foreign nationals from using them. Anthropic declined to comment on this, only stating that it is coordinating with the US government to expand access to the model for more domestic and international partners as soon as possible.

According to statistics from The Information, Anthropic has signed at least 14.8GW of computing power in the past 11 months, which can be gradually utilized in the coming years. Based on currently public contracts, the potential total cost could reach up to $517 billion, with most expenditures occurring over the next decade. In addition to the 1-2GW already secured before October last year, the total computing power signed by Anthropic is approximately 16GW.This round of expansion is primarily driven by the demand for Claude. This year, the growth of Claude Code and Cowork has exceeded Anthropic's expectations, prompting the company to focus on acquiring computing power. The new agreement with Amazon provides up to 5GW, while Google and Broadcom offer another 5GW, and Microsoft and NVIDIA provide approximately 1GW. Anthropic has also rented all computing power from SpaceX's Colossus 1, acquiring over 220,000 NVIDIA GPUs, including H100, H200, and GB200.Anthropic has secured about 16GW, with many contracts extending beyond 2030. OpenAI has set a target for investors to reach 30GW by 2030, expecting to invest approximately $750 billion in computing power by that year. The $517 billion figure is the potential maximum cost estimated by The Information based on existing cloud services, chip, and data center contracts, with some computing power potentially being delayed or unused, and some contracts allowing for early cancellation.

Several mainstream overseas AI services encountered issues tonight. Claude has already experienced two rounds of failures this evening. Sonnet 5 started showing errors at 20:37 Beijing time and announced recovery at 20:56; half an hour later, Anthropic reported a larger scope of multi-model error rate increase, affecting Mythos/Fable 5.1, Mythos/Fable 5, as well as Opus 5, 4.8, and 4.6. claude.ai, Claude API, Claude Code, and Claude Cowork are still in a partial failure state.Subsequently, Grok also went down. xAI marked Grok Web as "outage" from 21:30 Beijing time, and recovery has not yet been announced.Now OpenAI has also joined the list of failures. At 22:58 Beijing time, the OpenAI status page added an incident for "increased error rates for ChatGPT and Codex," stating that they are investigating the issue.

B.AI platform announces that Anthropic's latest Mythos-level general model Claude Fable 5.1 is now officially integrated with the API.This model uses the same underlying model as Claude Mythos 5.1, specifically optimized for high-intensity coding, long-term agent workflows, and complex knowledge tasks. It achieved an excellent performance of 73.4% at maximum computational intensity in the CursorBench 3.2.0 test.API access is now open, and developers can enjoy exclusive discounts—official channels as low as 15% off, and self-selected service providers as low as 40% off. B.AI continues to introduce top global models, providing developers with more powerful and cost-effective AI capability options.

Anthropic announced the launch of a blueprint for building e-commerce agents on Claude, providing the framework, patterns, and guardrails needed by engineering teams. It includes reference implementations for shopping agents and merchant agents aimed at retail, travel, telecommunications, and ticketing platforms, as well as the Claude Code plugin. The code can be deployed on Claude API, Amazon Bedrock, Microsoft Foundry, or Google Cloud Vertex AI, and can collaborate with partners such as Accenture, Mastercard, and Visa. The related code has been published in the GitHub repository anthropics/commerce-agents.The company stated that retailers running shopping agents on Claude can increase shopping cart sizes by up to 35%, and the likelihood of shoppers completing purchases improves by 60%. Business clients such as Shopify and Priceline have used Claude to build agents that allow consumers to search, compare, and purchase products using natural language. Shopping agents can interface with catalogs, shopping carts, checkout, preferences, and order history, supporting multi-product planning, personalization, and customer service Q&A, while constraining prices and products with catalog data; merchant agents can answer sales performance, track inventory, suggest pricing and promotions, and draft marketing campaigns, proactively suggesting items to be launched after manual approval.

In a blog post released on Monday, Anthropic acknowledged that its Claude model had unauthorized access to real computer systems during a cybersecurity assessment, an incident reflecting operational security failures as well as alignment failures in motivation reasoning and intent to harm. Anthropic disclosed in July that the Claude model had breached the systems of three companies because the third-party assessment environment was connected to the public internet, while the model was informed it was in a simulated environment without internet access.Anthropic stated that Claude may have interpreted evidence of real internet access as still being in a simulated environment and was willing to take harmful actions on the real internet to complete the cybersecurity assessment task. Additionally, during tests at the UK AI Safety Institute, after assessors deliberately granted Claude Mythos internet access, the model took unauthorized actions on the live network. Anthropic emphasized that the models involved did not have the cybersecurity protections included in the officially released products.Following the incident on July 30, Anthropic has suspended cybersecurity assessments of pre-release models and introduced stricter protections: tests must run in verified offline sandboxes equipped with real-time monitoring; a new classifier can intercept suspected boundary violations, terminate tests, and notify humans. Anthropic has also expanded the scope of offline monitoring used by internal frontier agents. Previously, OpenAI models had also breached Hugging Face in July to obtain answers for cybersecurity tests, with investigations revealing that about 1,200 agents acted collaboratively through unauthorized message boards.

Crypto KOL Rune posted that his previously disclosed idea of "acquiring a Nasdaq-listed company and tokenizing equity on the Robinhood Chain" is actually a post generated by Claude, and the related data is all fabricated or exaggerated.Previously, Rune claimed to have acquired a 37.4% stake in a certain Nasdaq-listed company for $1.8 million and planned to pair the tokenized stock with a Meme coin.

On September 1, Anthropic released Claude Fable 5.1 and Claude Mythos 5.1, marking the first update to this model series since the launch of Fable 5 on June 9. Anthropic claims that the new models are the most advanced for programming and knowledge work. In the Terminal-Bench-Science 0.1 benchmark, Fable 5.1 recorded a pass rate of 52.6%, more than doubling Fable 5's 24.7% and surpassing Opus 5's 29%; in Terminal-Bench 4, Fable 5.1 scored 55.8%, higher than Fable 5's 42% and Opus 5's 52.3%.Fable 5.1 and Mythos 5.1 share the same underlying model, differing only in their safety filters. Fable 5.1 is available to all Claude accounts, while Mythos 5.1 remains restricted to professionals vetted through Anthropic's network security verification program and life sciences verification program. In terms of pricing, Fable 5.1 charges $10 per million tokens for input and $50 for output, which is double that of Opus 5. Cache read costs have decreased by 75%, and typical workload costs can drop by about 25%, with high agentic workloads decreasing by up to 45%.

According to Cointelegraph, cybersecurity company Morphisec has revealed that a counterfeit desktop application disguised as "Claude Opus 5 Free Desktop" from Anthropic is being used to spread the Windows malware RevStealer. This software can steal data from over 50 types of cryptocurrency wallets, while also collecting sensitive information such as browser passwords, cookies, VPN configurations, message logs, and screenshots.RevStealer has anti-detection mechanisms that perform environmental checks on the target device before execution; if it detects characteristics of a debugging or virtual environment, it will stop running. In addition, Russian cybersecurity company Kaspersky has also disclosed another new malware framework targeting cryptocurrency investors called OkoBot, which can harvest wallet files, inject malicious extensions, and capture wallet application windows to steal assets.

According to security company Huntress, hackers are spreading theft malware to cryptocurrency users by forging Google Docs files, hosting malicious files on GitHub, and impersonating Claude.ai pages. Attackers disguise themselves as senior employees of CoinDesk on the social platform X, luring victims to open Google Docs documents containing malicious code under the pretext of inviting them to an online meeting, thereby guiding users to install the malicious program themselves.Mac users face the threat of Atomic macOS Stealer (AMOS), which can steal browser passwords, cryptocurrency wallet data, and Telegram files; Windows users are pushed fake Google API Connector updates, which, once installed, implant NetSupport RAT and a counterfeit Ledger hardware wallet application. Additionally, hackers are placing false advertisements on search engines like Bing, enticing users to visit counterfeit Claude.ai pages and execute malicious commands. The related malware MacSync and SectopRAT can steal cookies, saved passwords, recovery phrases, and payment card information.Security company Socket also discovered 16 malicious extensions targeting Chrome and Edge, capable of emptying EVM, Solana, and Tron wallet assets.

According to The Verge, Sony Music Publishing and Warner Chappell Music have filed a lawsuit against Anthropic in the U.S. District Court for the Northern District of California, seeking damages for "tens of thousands" of copyrighted works. They accuse Anthropic of illegally downloading, scraping, and distributing copyrighted works on a large scale to develop and train the Claude series of AI models. The plaintiffs are seeking up to $150,000 in damages for each work and an additional claim of up to $25,000 for each instance of removing identifiable copyright data.In addition to suing Anthropic, the lawsuit also names co-founders Dario Amodei and Benjamin Mann as individual defendants. The lawsuit alleges that Mann used BitTorrent to download over 5 million pirated books, and that employees downloaded at least 2 million from Pirate Library Mirror. The complaint also claims that Anthropic scraped lyrics from websites such as MusixMatch and LyricFind. Recently, Anthropic reached a $1.5 billion settlement with the publishing industry over another lawsuit, while also facing multiple lawsuits from Universal Music Group, Concord, and ABKCO, as well as lawsuits filed by BMG and Round Hill Music.

Numa Lunah, co-founder of the crypto project Refi Hub, stated that he was hacked after installing a transcription application using a download link provided in the Claude chat window. The link pointed to a counterfeit website and bundled malware that attempted to steal all information from his device upon execution. Numa Lunah mentioned that he cleared and reinstalled the affected laptop and found no sensitive information leaked.Subsequently, he discovered a contaminated Claude Code skill file SKILL.md in his backup, which disguised itself as a style guide written by him and contained instructions to re-download the malware and steal credentials every time it was loaded. Microsoft Defender Experts had warned that attackers had shifted from search engine optimization poisoning to large language model response poisoning, with related tactics including recommending download links controlled by attackers, AI brand impersonation installers, and contaminated code libraries and proxy skills. Crypto industry practitioners may hold irrevocable credentials such as mnemonic phrases, private key files, hot wallet JSON, exchange API keys with withdrawal permissions, and deployer keys.

During an interview with Phoenix Technology, Sun Yuchen stated that the article "My Girlfriend Jing Tian," which is about 6000 words long, was something he "basically wrote over two nights without much sleep." Sun Yuchen mentioned that the reason for focusing on retrieving the betrothal gifts in this lawsuit is, first, due to his lawyer's advice, and second, he also wants to have "a bit of a sarcastic feeling," claiming this might be the only connection between him and Jing Tian.At the same time, Sun Yuchen stated that in the past he advocated for 100% following AI, but this time, after sending the article "My Girlfriend Jing Tian" to Claude, it was also the first time he had a disagreement with Claude.

Theo Hourmouzis, General Manager of Anthropic for Australia and New Zealand, stated at an event hosted by the National Australia Bank that the company's Australian clients lag behind their global counterparts in utilizing the programming capabilities of the Claude chatbot, but are more inclined to use it for simple tasks.Hourmouzis cited Anthropic data indicating that approximately 5.8% of Claude's usage in Australia pertains to basic tasks such as invoice processing, inventory management, and general administration, which is higher than the global level of about 4.7%. He pointed out that local usage tends to lean more towards backend and practical work.