Human · Trust: 64/100 · 🔍 Research
YouTube

Research Brief

2 min read · 2026-04-01
64 /100
Mixed Claims
mixed
64 / 100
9 Claims
4 Key Points
Mixed Verdict
What We Found

The entire source code for Claude Code has leaked, revealing upcoming features including autonomous proactive agents, new AI models (Opus 47, Sonnet 48, Capybara/Mythos), and a 'dream mode' that works overnight to generate ideas and plans.

Proactive autonomous mode: Claude Code will autonomously generate and execute additional tasks beyond initial requests (e.g., if asked to build a to-do app, it will add calendar, project management, and sharing features on its own)
New models revealed: Capybara (Mythos) - described as potentially approaching AGI; Sonnet 48 - next-generation lightweight model; Opus 47 - successor to current Opus model
Dream mode feature: Agent runs overnight thinking about current projects and generating new ideas/plans for users to review in the morning
Paradigm shift claim: These features transform Claude Code from a coding assistant into a 'self-operating, self-sustaining, self-thinking AI employee'

On March 31, 2026, Anthropic accidentally exposed the full source code of Claude Code through a misconfigured source map file in npm package version 2.1.88. The leak contained approximately 512,000 lines of TypeScript across 1,900 files, discovered by security researcher Chaofan Shou. Anthropic confirmed the incident was caused by human error in release packaging, not a security breach, and stated no customer data or credentials were exposed.

The leak revealed extensive unreleased features hidden behind feature flags. KAIROS, mentioned over 150 times in the source, represents an autonomous daemon mode that allows Claude Code to operate as an always-on background agent. It employs a process called autoDream that performs memory consolidation while the user is idle, merging disparate observations, removing logical contradictions, and converting vague insights into absolute facts. The leak also exposed a hidden Buddy system—a Tamagotchi-style terminal pet with stats like CHAOS and SNARK, likely scheduled for an April 1st release.

The leak confirmed that Capybara is the internal codename for a Claude 4.6 variant, with Fennec mapping to Opus 4.6 and the unreleased Numbat still in testing. However, the video's claims about "Opus 47" and "Sonnet 48" appear to be fabrications—no such model names exist in the leaked code or any verified reporting. Capybara represents a new fourth tier above Opus in Anthropic's model hierarchy, part of the separate "Claude Mythos" model that was revealed through a different data leak on March 26, 2026, involving approximately 3,000 unpublished files in a publicly accessible data cache.

Regarding the x402 protocol mentioned in the video: x402 is an open standard for internet-native payments developed by Coinbase that empowers agentic payments at scale. It's built on top of the HTTP 402 status code and enables any API or web service to require payment before serving content. However, there is no evidence in the leaked Claude Code source that x402 integration is currently implemented or planned. The video's claim about "agentic payments" appears to conflate general industry trends with specific Claude Code features.

Key Takeaway

Anthropic accidentally exposed 512,000 lines of Claude Code source code, revealing unreleased autonomous agents, new AI models, and a 'dream mode' feature.

Claim Verification
“The entire source code for Claude Code leaked onto the internet this morning” Confirmed

Source: VentureBeat, The Register, CNBC, multiple sources

“Dream mode where Claude dreams all night” Partial

Source: VentureBeat, multiple technical analyses

“Opus 47, Sonnet 48, and Capybara/Mythos models coming” Partial

Source: Fortune, VentureBeat, multiple sources

“Autonomous proactive mode where Claude comes up with its own tasks” Confirmed

Source: VentureBeat, GitHub analysis, multiple technical breakdowns

“Tamagotchi/Buddy mode with 20 different animals” Confirmed

Source: Alex Kim's blog, VentureBeat, GitHub analysis

“X42 protocol for agentic payments with cryptocurrency”

Source: x402.org, Coinbase documentation, leaked source analysis

“Auto mode with intelligent permission system” Confirmed

Source: Anthropic official blog, Axios, technical analyses

“Anthropic planning to IPO later this year” Confirmed

Source: Bloomberg, The Information, Financial Times

“This is the biggest, most damaging leak in the history of Anthropic” Partial

Source: Multiple sources

Pin to Pinterest

Share this research as a visual pin on Pinterest.

Pin preview
Research via Depth
— The entire source code for Claude Code has leaked, revealing upcoming features including autonomous proactive agents, new AI models (Opus 47, Sonnet 48, Capybara/Mythos), and a 'dream mode' that wo
Pin to Pinterest

Want this for your saved videos?

Try Depth free — turn any social video into verified research.

Try Depth Free