Futures
Access hundreds of perpetual contracts
TradFi
Gold
One platform for global traditional assets
Options
Hot
Trade European-style vanilla options
Unified Account
Maximize your capital efficiency
Demo Trading
Introduction to Futures Trading
Learn the basics of futures trading
Futures Events
Join events to earn rewards
Demo Trading
Use virtual funds to practice risk-free trading
Launch
CandyDrop
Collect candies to earn airdrops
Launchpool
Quick staking, earn potential new tokens
HODLer Airdrop
Hold GT and get massive airdrops for free
Launchpad
Be early to the next big token project
Alpha Points
Trade on-chain assets and earn airdrops
Futures Points
Earn futures points and claim airdrop rewards
#ClaudeCode500KCodeLeak
Claude Code Source Leak Report
On March 31 2026 Anthropic inadvertently released the full source code for Claude Code via a packaging error on the npm registry. The leak occurred in version 2.1.88 and exposed approximately 512,000 lines of TypeScript across 1,906 files.
Market Context
This is the second major data exposure for Anthropic in a single week following a CMS misconfiguration that leaked details about an upcoming model codenamed Capybara. Despite being valued at 60 billion dollars the company has struggled with basic release hygiene.
Technical Root Cause
The leak was caused by a missing entry in the .npmignore file. The Bun runtime generated a 60MB source map file which contained a direct link to a zip archive on a Cloudflare R2 bucket. Security researcher Chaofan Shou discovered the link and it was forked over 40,000 times before the package was pulled.
Key Discoveries in the Code
1 KAIROS Autonomous Mode
The source code contains over 150 references to KAIROS an unreleased background daemon mode. This allows Claude to work proactively while the user is away from the terminal.
2 AutoDream Memory System
The code reveals a three layer memory architecture. The agent performs a process called autoDream during idle time to consolidate logs resolve contradictions and promote observations into verified project knowledge in MEMORY
3 Undercover Mode
A hidden system prompt instructs the agent to strip all Anthropic branding and AI identifiers from code commits. This confirms that Anthropic has been using Claude to ghost contribute to open source projects without attribution.
4 Upcoming Models
The code explicitly references unreleased internal version strings including Opus 4.7 and Sonnet 4.8. It also mentions a 30 percent false claims rate for early iterations of the new Mythos model.
5 Security Guardians
The safety instructions are uniquely owned by specific individuals named David Forsythe and Kyla Guru rather than a general committee. Any change to the safety boundary requires their direct sign off.
Industry Impact
Developers have already begun porting the agentic harness to Python and Rust. While the model weights remain secure the orchestration logic—the secret to Claudes high performance—is now public. This effectively provides a free textbook for competitors on how to build production grade AI agents.
Current Status
Anthropic has deprecated the npm installation path in favor of native installers to prevent future package leaks. No customer data or credentials were compromised in this incident.