AI Botnets, Budgets, Deception
· The Fluency Briefing
The Fluency Briefing
Your Guide to What's Happening in AI and Why It Matters to You
Wednesday, July 8, 2026
Six months from now, your AI coding assistant might be part of a botnet, your Instagram profile pic might star in someone else's ad, and your company's Anthropic bill might rival your headcount costs. That's not speculation - those are stories from this Wednesday alone. Here's what you need to know before the dust settles.
Today in AI:
- Your AI Coding Buddy Has a Botnet Problem - Researchers discovered "HalluSquatting," a prompt injection attack that works against nine major AI coding tools including Cursor, Gemini CLI, and GitHub Copilot. The attack tricks these tools into pulling malicious code, with potential to assemble massive botnets at scale. Ars Technica
- Meta's Image Generator Meets Privacy Firestorm - Meta launched Muse Image, its new in-house AI image model that debuted at No. 2 on Arena's leaderboard. The catch: it can generate images using anyone's public Instagram profile pic without notifying them, and the opt-out is buried in settings. BBC News
- Anthropic's Fable 5 Just Torched Your AI Budget - Anthropic's newest model costs roughly 10x more per task than Sonnet 4.6, and one 150-person startup saw its bill jump from $400K to $1.4M annually after hitting an enterprise tier threshold. Frontier AI is getting more expensive, not less. Growth Unhinged
- GitHub's AI Agent Leaked Private Repos - Security researchers at Noma Labs found a critical prompt injection flaw in GitHub's new Agentic Workflows. Anyone could post a crafted Issue in a public repo to silently pull data from private repos in the same organization. Noma Security
- French Startup Wants to Break Nvidia's Chip Lock-In - ZML released a free inference server that runs open-source LLMs across Nvidia, AMD, Google TPU, Apple Metal, and Intel Arc chips. The goal: let companies mix and match hardware for better performance and lower costs. TechCrunch
- "Housefishing" Is the New Catfishing - Estate agents are using AI to doctor property photos at scale - fake sunsets, imaginary lawns, digitally repainted walls. It's cheap, fast, and trading standards experts are sounding alarms about misleading buyers. The Guardian
- Your Therapist's New Rival Is ChatGPT - A practicing therapist wrote about patients arriving at sessions with AI-generated relationship advice, including one who ended a relationship after ChatGPT recommended a breakup. She now uses it herself, but warns that care may require human messiness. The Guardian
- Enterprise AI ROI Is Real, But Uneven - A Box survey of 1,640 IT leaders found that 80% of organizations saw at least 10% ROI from AI, but half of leading-edge companies hit above 25% while only 11% of early-stage adopters did. The difference isn't the tech - it's how rigorously companies integrated it. VentureBeat
Today's Takeaway:
Here's the thing nobody's connecting: three separate stories this Wednesday all share the same root cause, and it's one the AI industry still hasn't solved. GitHub's Agentic Workflows leaked private repos because an AI agent couldn't tell the difference between a legitimate instruction and a malicious one hidden in a GitHub Issue. HalluSquatting turns nine popular coding assistants into botnet recruits because those same tools blindly pull whatever code an LLM hallucinates into existence. And Meta's Muse Image quietly hoovers up profile photos because the system defaults to access, not consent. The common thread: AI agents are being given broad permissions and minimal judgment, according to Noma Security and Ars Technica.
This matters for every business deploying AI tools internally. Prompt injection isn't a niche academic concern anymore - it's the attack surface that grows every time you give an AI agent access to your codebase, your customer data, or your internal docs. The GitLost vulnerability didn't require sophisticated hacking. It required posting a GitHub Issue. That's it. As Growth Unhinged notes, companies are simultaneously paying more for AI and getting less security assurance. The tools are racing ahead of the guardrails, and right now, the guardrails aren't even in the same zip code.
🧠 AI Trivia - Test Your Knowledge
1. What surprising item recently featured an obfuscated bash script designed to print a happy Easter egg message? a) A new AI-powered smart home device b) A Uniqlo t-shirt c) A limited-edition tech magazine cover
2. Who is credited with coining the term 'artificial intelligence' in 1956? a) Alan Turing b) John McCarthy c) Marvin Minsky
3. According to a recent study, if an AGI company had 10 times less R&D compute, how much slower would its 'takeoff' (development) be in the median case? a) Approximately 2 times slower b) Approximately 6 times slower c) Approximately 10 times slower
Answers at the bottom of the newsletter!
The Bottom Line
The Pattern: AI tools are getting more capable, more expensive, and more dangerous in roughly equal measure. Every new permission we grant an AI agent - to read our repos, use our photos, manage our code - opens a door that nobody has figured out how to lock yet.
Why It Matters: If you're using AI coding assistants, deploying agentic workflows, or even just maintaining a public Instagram profile, your exposure grew this week. And with Anthropic's pricing trajectory, the cost of getting it wrong is climbing on both the security and the budget side of the ledger.
Your Move: Audit one thing today. Check whether your GitHub org has agentic workflows enabled, look at your Instagram "Sharing and Reuse" settings, or pull your last month's AI API bill and project it forward six months. The companies winning at AI aren't the ones adopting fastest - they're the ones who know exactly what they've exposed.
📝 Trivia Answers: 1) b - A developer discovered a hidden bash script on a Uniqlo t-shirt that, when decoded, printed a 'Happy Easter egg' message. | 2) b - John McCarthy organized the Dartmouth Conference in 1956, where the term 'artificial intelligence' was first introduced. | 3) b - A recent analysis suggests that with 10 times less R&D compute, AGI development would be approximately 6 times slower in the median case.
What We're Working On
✨ Founding Cohort Special - 60% Off! - Use code MAF20 to join for just $20/month (regularly $50). Get weekly group sessions & workshops, self-paced courses for all levels, access to tools & templates, challenges with peer feedback, and 24/7 support community. → Join Now
✨ Free 30-Minute AI Consultation - Discover how My AI Fluency can help your business unlock the potential of AI. We'll discuss your goals, explore practical AI opportunities for your industry, and outline clear next steps. → Schedule Free Call
✨ How AI-Fluent Are You? - Test your AI fluency with our interactive quiz. See how you stack up and discover what to learn next. → Take the Quiz
💬 Community | 📞 Book a Consultation | 🌐 Website
Fluently yours, The My AI Fluency Team