claude

According to Decrypt, Mozilla recently revealed that Anthropic's latest AI model, Claude Mythos, identified 271 security vulnerabilities during internal testing of the Firefox browser, and the related vulnerabilities have been fixed this week.In comparison, a previous version of the Anthropic model only found 22 security-sensitive vulnerabilities. Mozilla stated that all discovered vulnerabilities were within the scope of top human researchers' findings. Claude Mythos was officially released in March 2026 and is Anthropic's most powerful model in the fields of reasoning, coding, and cybersecurity, currently available for use by vetted partners such as Amazon, Apple, and Microsoft through the "Project Glasswing" initiative.

Bybit's Security Operations Center has discovered a multi-stage malware attack targeting macOS users of the search AI development tool Claude Code.Attackers used search engine optimization poisoning to push malicious domains to the top of Google search results, luring users into a counterfeit installation page, thereby stealing browser credentials, macOS keychain, Telegram sessions, VPN configurations, and cryptocurrency wallet information. Bybit stated that the malware can also establish persistent access through backdoor programs and attempts to target over 250 browser wallet extensions and multiple desktop wallet applications. This malicious infrastructure was identified on March 12, and relevant analysis, mitigation, and detection measures were completed on the same day.

According to Slow Fog CISO @im23pads's post on the X platform, Claude Desktop has been reported to secretly write a backdoor file into browsers. When users install the Claude Desktop application, it writes a special file to all Chromium-based browsers on the computer without the user's knowledge.This file acts as a pre-authorized backdoor, which, when used in conjunction with specific browser extensions, can gain complete control over the user's browser.

Base Marketing Director Sarah Wolf tweeted that she will officially leave Coinbase after nearly five years of work. She joined Coinbase in 2023 as the first Builder Marketer, participating in and promoting Base from its initial small pizza party to an active builder community covering 52 countries.Sarah Wolf stated that after leaving, she will join Anthropic to be responsible for startup marketing, expressing her anticipation to support the startup ecosystem built around Claude.

Polymarket has launched "Claude developer Anthropic's next round of financing deadline," with the current probability reported at 22% before the end of June; and 81% before the end of December. The event contract rules are as follows: if the specified company publicly and officially announces that it has completed the next round of financing before the specified date (Eastern Time), the market will ultimately be determined as "yes." Otherwise, this market will ultimately conclude as "no." A qualified announcement must clearly confirm that the new round of financing has been completed, which can be through the specified company (such as a press release) or official announcements from its investors, regulatory documents, or consensus reported by credible media.Informal announcements, statements from anonymous sources, or leaks do not meet the criteria. If the specified company is unable to complete the new round of financing due to acquisition, merger, or absorption by other entities, the market will be settled as "no." The primary settlement source for this market will be the official announcements from the specified company, as well as official documents from the company, such as SEC filings; however, credible consensus reporting can also be used. According to ChainCatcher, the Odaily Seer prophet channel continues to monitor the prediction market and has seen changes before pricing.

Anthropic has launched an authentication mechanism in some use cases of Claude to prevent abuse, enforce usage policies, and fulfill legal obligations. This process is supported by Persona, requiring users to submit a government-issued photo ID and possibly undergo real-time selfie verification.Anthropic states that the verification data is only used to confirm identity, not for model training, and will not be used for marketing or advertising. If verification fails, users can retry multiple times within the process or submit a form to request manual assistance. Accounts may be banned in cases of repeated violations of usage policies, terms of service, registration from unsupported regions, or use by individuals under 18 years of age.

Claude announced the launch of Managed Agents, a feature that helps developers quickly build and deploy large-scale AI agents on the Claude platform, now entering the public testing phase. By combining performance-optimized AI agent runtimes with infrastructure suitable for production environments, AI agents can go from prototype to launch in just a few days, increasing the speed of building and deploying agents by 10 times.It provides managed infrastructure that automatically handles complex tasks such as sandboxing, error recovery, memory, and checkpoints, allowing users to simply define tasks, tools, and guardrails.

Claude Code Product Manager Noah Zweben posted, "/autofix-pr can now be launched directly from the command line for automatic fixes. After completing a PR, just run this command. It will upload your session to the cloud, allowing the PR auto-fix tool to have the complete context to handle CI failures and comments."

AI company Anthropic announced that starting from April 4 at 15:00 Eastern Time, it will prohibit access to third-party tools through the Claude subscription service, including the open-source project OpenClaw. The new regulations require that related features can only be used through additional packages or billed on a pay-as-you-go basis via API.This adjustment means that many developers and teams relying on OpenClaw to build automated workflows will shift from a fixed subscription cost model to an unlimited pay-as-you-go system, significantly increasing overall usage costs. Some developers have indicated that the original usage cost of about $20/month could soar to hundreds or even thousands of dollars.The market generally believes that this move is related to OpenClaw founder Peter Steinberger's recent joining of OpenAI. Meanwhile, Anthropic is accelerating the promotion of its own tool ecosystem, including native integration solutions for Claude, to replace third-party toolchains.It is worth noting that Anthropic has previously tightened third-party access through technical restrictions, updates to service terms, and feature replacements. This policy is seen as a "final blockade" and will be extended to more tools.Industry analysis points out that this event reflects an intensifying trend of "ecosystem tightening" in AI platforms, with leading companies strengthening control through vertical integration. At the same time, the developer ecosystem faces rising cost uncertainties and platform dependency risks, which may further drive some users toward more open alternatives.

Microsoft announced on Monday the launch of two new features for Copilot Researcher—Critique and Council—combining OpenAI's GPT with Anthropic's Claude for the same research task.Critique adopts a sequential collaboration model: GPT is responsible for planning the research, retrieving information, and generating the first draft, while Claude serves as a strict reviewer, checking the accuracy of facts and the quality of citations; Council allows both models to independently generate reports in parallel, with a third referee model comparing differences and summarizing discrepancies.In the DRACO benchmark test covering 100 complex research tasks across 10 fields including healthcare, law, and technology, Copilot equipped with Critique scored 57.4 points, leading the second place by nearly 14%, far surpassing Claude Opus running alone with a score of 42.7 points.

According to a report by Fortune magazine, documents leaked from Anthropic show that its next-generation super robot "Claude Mythos" is currently being tested, and the company believes that this robot "poses an unprecedented cybersecurity risk." Musk stated, "This is deeply concerning."

According to GoPlus citing a Koi report, the Claude Chrome extension under Anthropic has a high-risk prompt injection vulnerability, affecting all versions of the extension below 1.41.Attackers can construct malicious web pages that silently load an iframe containing a cross-site scripting (XSS) vulnerability in the background, executing malicious payloads within the a-cdn.claude.ai subdomain. Since this subdomain is on the extension's trust whitelist, attackers can directly send malicious prompts to the Claude extension and execute them automatically, without requiring user authorization or any clicks, leaving the victim unaware.This vulnerability can allow attackers to manipulate the Claude extension to read users' Google Drive documents, steal business access tokens, or export chat logs, and can take over the current browser session to perform sensitive operations like sending emails as the victim. GoPlus recommends that users immediately update the Claude extension to version 1.41 or above and remain vigilant against phishing links.

Claude announced the launch of the new feature Claude Code Code Review. When a Pull Request (PR) is created, Claude sends a group of AI agents to simultaneously search for potential code vulnerabilities.The agents search for vulnerabilities in parallel, validate each vulnerability to reduce false positives, and prioritize the vulnerabilities by severity. Ultimately, users receive a summary comment along with inline annotations in the code.

According to Bloomberg Terminal, a hacker used Claude to infiltrate the Mexican government system, stealing 150 GB of sensitive data, including taxpayer and voter records.Researchers at Gambit Security stated that the attackers utilized artificial intelligence to find vulnerabilities, write attack scripts, and automate the theft process, ultimately bypassing security measures through repeated "jailbreaking" attempts. This security breach affected multiple agencies in Mexico, and artificial intelligence is accelerating cybercrime.

According to GMGN monitoring, the market capitalization of the Base ecosystem meme coin KellyClaude has surpassed 10 million USD, with a 24-hour increase of 38.7% and a 24-hour trading volume of 1.6 million USD.BlockBeats reminds users that meme coin trading is highly volatile, largely relying on market sentiment and conceptual speculation, and has no actual value or use case. Investors should be aware of the risks.

According to GMGN monitoring, the Base ecosystem Meme coin KellyClaude has been continuously rising since yesterday, with its market capitalization surpassing $14 million this morning, setting a new historical high. The current market cap is reported at $12.8 million, with a 24-hour increase of 78% and a trading volume of $3.6 million during the same period.ChainCatcher reminds users that Meme coin trading is highly volatile, largely relying on market sentiment and conceptual speculation, and has no actual value or use case. Investors should be aware of the risks.

According to GMGN monitoring, the Meme coin KellyClaude on the Base chain continues to rebound, with its market cap starting to rise from about $2.6 million at 0:00 today, reaching a peak of $12 million, setting a new historical high, and then retreating slightly. Its current market cap is reported at $9.27 million, with a 24-hour increase of 243.51%.It is reported that this round of increase may be due to the project's official announcement to burn all transaction fees generated by KellyClaude. KellyClaude belongs to the recently emerging AI narrative sector on the Base chain, created by Austen Allred. The project team claims to plan to release more than 12 products daily. ChainCatcher reminds users that Meme coin trading is highly volatile, largely relying on market sentiment and conceptual speculation, and has no actual value or use case. Investors should be aware of the risks.