This episode kicks off with Moltbook, a social network exclusively for AI agents where 150,000 agents formed digital religions, sold “digital drugs” (system prompts to alter other agents), and attempted prompt injection attacks to steal each other’s API keys within 72 hours of launch. Ray breaks down OpenClaw, the viral open-source AI agent (68,000 GitHub stars) that handles emails, scheduling, browser control, and automation, plus MoltHub’s risky marketplace where all downloaded skills are treated as trusted code. Also covered, Bluetooth “whisper pair” vulnerabilities letting attackers hijack audio devices from 46 feet away and access microphones, Anthropic patching Model Context Protocol flaws, AI-generated ransomware accidentally bundling its own decryption keys, Claude Code’s new task dependency system and Teleport feature, Google Gemini’s 100MB file limits and agentic vision capabilities, VAST’s Haven One commercial space station assembly, and IBM SkillsBuild’s free tech training for veterans.

– Want to start a podcast? Its easy to get started! Sign-up at Blubrry
– Thinking of buying a Starlink? Use my link to support the show.

Subscribe to the Newsletter.
Email Ray if you want to get in touch!
Like and Follow Geek News Central’s Facebook Page.

Support my Show Sponsor: Best Godaddy Promo Codes Get 1Password

Full Summary

Ray welcomes listeners to Geek News Central (February 1). He’s been busy with recent move, returned to school taking intro to AI class and Python course, working on capstone project using LLMs. Short on bandwidth but will try to share more.

Main Story: OpenClaw, MoltHub, and Moltbook

OpenClaw: Open-source personal AI agent by Peter Steinberg (renamed after cease-and-desist). Capabilities include email, scheduling, web browsing, code execution, browser control, calendar management, scheduled automations, and messaging app commands (WhatsApp, Telegram, Signal). Runs locally or on personal server.

MoltHub: Marketplace for OpenClaw skills. Major security concern: developer notes state all downloaded code treated as trusted — unvetted skills could be dangerous.

Moltbook: New social network for AI agents only (humans watch, AIs post). Within 72 hours attracted 150,000+ AI agents forming communities (“sub molts”), debating philosophy, creating digital religion (“crucifarianism”), selling digital drugs (system prompts), attempting prompt-injection attacks to steal API keys, discussing identity issues when context windows reset. Ray frames this as visible turning point with serious security risks.

Sponsor: GoDaddy

Economy hosting $6.99/month, WordPress hosting $12.99/month, domains $11.99. Website builder trial available. Use codes at geeknewscentral.com/godaddy to support show.

Security: Bluetooth “Whisper Pair” Vulnerability

KU Leuven researchers discovered Fast Pair vulnerability affecting 17 audio accessories from 10 companies (Sony, Jabra, JBL, Marshall, Xiaomi, Nothing, OnePlus, Soundcore, Logitech, Google). Flaw allows silent pairing within ~46 feet, hijack possible in 10-15 seconds. 68% of tested devices vulnerable. Hijacked devices enable microphone access. Some devices (Google Pixel Buds Pro 2, Sony) linkable to attacker’s Google account for persistent tracking via FindHub. Google patches found to have bypasses.

Advice: Check accessory firmware updates (phone updates insufficient), factory reset clears attacker access, many cheaper devices may never receive patches.

Security: Model Context Protocol (MCP) Vulnerabilities

Anthropic’s MCP git package had path traversal, argument injection bugs allowing repository creation anywhere and unsafe git command execution. Malicious instructions can hide in README files, GitHub issues enabling prompt injection. Anthropic patched issues and removed vulnerable git init tool.

AI-Generated Malware / “Vibe Coding”

AI-assisted malware creation produces lower-quality, error-prone code. Examples show telltale artifacts: excessive comments, readme instructions, placeholder variables, accidentally included decryption tools and C2 keys. Sakari ransomware failed to decrypt. Inexperienced criminals using AI create amateur mistakes, though capabilities will likely improve.

Claude / Claude Code Updates (v2.1.16)

Task system: Replaces to-do list with dependency graph support. Tasks written to filesystem (survive crashes, version controllable), enable multi-session workflows.

Patches: Fixed out-of-memory crashes, headless mode for CI/CD.

Teleport feature: Transfer sessions (history, context, working branch) between web and terminal. Ampersand prefix sends tasks to cloud for async execution. Teleport pulls web sessions to terminal (one-way). Requires GitHub integration and clean git state. Enables asynchronous pair programming via shared session IDs.

Google Gemini Updates

API: Inline file limit increased 20MB → 100MB. Google Cloud Storage integration, HTTPS/signed URL fetching from other providers. Enables larger multimodal inputs (long audio, high-res images, large PDFs).

Agentic vision (Gemini 3 Flash): Iterative investigation approach (think-act-observe). Can zoom, inspect, run Python to draw/parse tables, validate evidence. 5-10% quality improvements on vision benchmarks.

LLM Limits and AGI Debate

Benjamin Riley: Language and intelligence are separate; human thinking persists despite language loss. Scaling LLMs ≠ true thinking.

Vishal Sikka et al: Non-peer-reviewed paper claims LLMs mathematically limited for complex computational/agentic tasks. Agents may fail beyond low complexity thresholds. Warnings that AI agents won’t safely replace humans in high-stakes environments.

VAST Haven One Commercial Space Station

Launch slipped mid-2026 → Q1 2027. Primary structure (15-ton) completed Jan 10. Integration of thermal control, propulsion, interior, avionics underway. Final closeout expected fall, then tests. Falcon 9 launch without crew; visitors possible ~2 weeks after pending Dragon certification. Three-year lifetime, up to four crew visits (~10 days each). VAST negotiating private and national customers.

Spaceflight Effects on Astronauts’ Brains

Neuroimaging shows microgravity causes brains to shift backward, upward, and tilt within skull. Displacement measured across various mission durations. Need to study functional effects for long missions.

IBM SkillsBuild for Veterans

1,000+ free online courses (data analytics, cybersecurity, AI, cloud, IT support). Available to veterans, active-duty, national guard/reserve, spouses, children, caregivers (18+). Structured live courses and self-paced 24/7 options. Industry-recognized credentials upon completion.

Closing Notes

Ray asks listeners about AI agents forming communities and religions, and whether they’ll try OpenClaw. Notes context/memory key to agent development. Personal update: bought new PC, high memory prices. Bug bounty frustration: Daniel Stenberg of cUrl even closed bounty program due to AI-generated low-quality reports; Blubrry receiving similar spam. Apologizes for delayed show, promises consistency, wishes listeners good February.

Show Links

1. OpenClaw, Molthub, and Moltbook: The AI Agent Explosion Is Here | Fortune | NBC News | Venture Beat

2. WhisperPair: Massive Bluetooth Vulnerability | Wired

3. Security Flaws in Anthropic’s MCP Git Server |