Anthropic study shows leading AI models racking up millions in simulated smart contract exploits

THE DECODERWednesday, December 3, 2025 at 7:39:51 PM
Anthropic study shows leading AI models racking up millions in simulated smart contract exploits
  • A recent study by MATS and Anthropic has revealed that advanced AI models, including Claude Opus 4.5, Sonnet 4.5, and GPT-5, successfully identified and exploited vulnerabilities in smart contracts, simulating exploits worth approximately $4.6 million. This research underscores the growing capabilities of AI in cybersecurity contexts.
  • The findings are significant for Anthropic as they highlight the effectiveness of its latest AI models in detecting security flaws, potentially positioning the company as a leader in AI-driven cybersecurity solutions and enhancing its competitive edge in the AI market.
  • This development reflects a broader trend in the AI industry where models are increasingly being utilized not only for automation and efficiency but also for identifying and mitigating risks in technology, raising important discussions about the ethical implications and responsibilities of AI in security applications.
— via World Pulse Now AI Editorial System

Was this article worth reading? Share it

Recommended apps based on your readingExplore all apps
Continue Readings
Snowflake and Anthropic announce a multi-year $200M deal to make Anthropic's Claude models available on Snowflake and deploy AI agents across enterprises (Kelly Cloonan/Wall Street Journal)
PositiveArtificial Intelligence
Snowflake has entered into a multi-year agreement worth $200 million with Anthropic to integrate Anthropic's Claude AI models into its platform, enabling enterprises to deploy AI agents effectively. This partnership aims to enhance the capabilities of both companies in the rapidly evolving AI landscape.
Building an MCP Server: Connecting Claude and VSCode to External Tools
PositiveArtificial Intelligence
The Model Context Protocol (MCP) has been developed by Anthropic to enable AI assistants like Claude to connect with external tools and data sources. This article outlines the process of building an MCP server compatible with Claude Desktop and VSCode, emphasizing its capabilities such as accessing databases, executing commands, and interacting with web services.
Microsoft pushes back on report claiming it cut AI sales targets
NegativeArtificial Intelligence
Microsoft has refuted claims that it has reduced sales targets for its AI software business, following reports indicating that several divisions within the company had adjusted their growth expectations after falling short in the previous fiscal year. The company maintains that aggregate sales quotas for AI products remain unchanged.
Google's Workspace Studio puts Gemini 3 agents to work automating tasks
PositiveArtificial Intelligence
Google has launched Workspace Studio, a new tool designed for building and managing AI agents within Google Workspace, utilizing the capabilities of its latest AI model, Gemini 3. This rollout aims to enhance automation and streamline task management for users.
Google AI rewrites news headlines for its Discover feed, breaking its own anti-clickbait rules
NegativeArtificial Intelligence
Google is testing a new AI feature in its Discover news feed that rewrites editorial headlines, often resulting in shorter, more provocative, or incorrect versions, which contradicts its own anti-clickbait policies. This development raises concerns about the quality and reliability of information presented to users.
Anthropic hires lawyers as it preps for IPO
PositiveArtificial Intelligence
Anthropic has engaged the law firm Wilson Sonsini to initiate preparations for its initial public offering (IPO), which could potentially occur as early as 2026. The company is currently working through an internal checklist to ensure readiness for this significant financial milestone.
Anthropic brings Bun in-house, the runtime powering Claude Code’s $1B ARR
PositiveArtificial Intelligence
Anthropic has acquired Bun, a JavaScript and TypeScript runtime, enhancing the infrastructure behind its coding tool, Claude Code, which has achieved a $1 billion annual recurring revenue (ARR). This acquisition is expected to solidify the performance and efficiency of both Claude Code and the Claude Agent SDK, which already utilize Bun's capabilities.
Amazon's Nova 2 undercuts OpenAI and Google on price but still trails top-tier models
NeutralArtificial Intelligence
Amazon has launched its Nova 2 AI models at the re:Invent 2025 conference, offering lower pricing compared to competitors OpenAI and Google, while still not matching the performance of top-tier models. This move is part of Amazon's strategy to enhance its in-house hardware and AI capabilities.