Dakarda AI Newsletter · 13 July 2026
Monday Edition
Apple Sues OpenAI Over Trade Secrets
Apple has filed a federal lawsuit against OpenAI, alleging systematic theft of trade secrets — from engineers to the Chief Hardware Officer. This is the first such bitter conflict between a consumer hardware giant and an AI leader, and it could change the rules of the game in the AI device market.
Intro · Alex
In the world of AI, it's rare to see such a plot twist that makes you rub your eyes. Apple — a company that only recently integrated ChatGPT into its systems — is suing OpenAI across 41 pages of indictment, alleging systematic theft of trade secrets. Over 400 former Apple employees now work at OpenAI, and the lawsuit describes misconduct at every level: from engineers to a former VP of Hardware, who allegedly encouraged candidates to bring 'actual Apple hardware components' to interviews. In this edition, I also tracked the second extension of Fable 5, which shows that even Anthropic has infrastructure problems; surprising data showing that one in four companies is putting AI agents on organizational charts; and funding rounds that prove the market for local models and dedicated AI hardware isn't slowing down. Plus, a warning from Ars Technica about a new type of attack that uses AI assistant hallucinations to build botnets — this is really worth knowing.
What's worth knowing
Apple Sues OpenAI — Systematic Theft of Trade Secrets
Apple has filed a 41-page federal lawsuit against OpenAI, alleging systematic theft of trade secrets and breach of contract at every level — from engineers to the Chief Hardware Officer. OpenAI allegedly encouraged Apple candidates to bring 'actual hardware components' to job interviews. Over 400 former Apple employees now work at OpenAI. This is the first major legal conflict between a consumer hardware giant and an AI leader, which could impact the entire AI device market.
Anthropic Delays Fable 5 Again — Second Time in 6 Days
Just hours after the previous deadline expired, Anthropic announced a second extension of access to Claude Fable 5 across all paid plans — until July 19. The 50% higher weekly limits in Claude Code were also maintained. The company cites compute constraints and the need to better estimate demand. For users, this means week-by-week uncertainty about access to the flagship model, and for the market, it signals that Anthropic has an infrastructure capacity problem.
23% of Companies Are Listing AI Agents on Org Charts
According to a Korn Ferry survey of 1,200 managers, 23% of companies formally place AI agents on their organizational charts, treating them like employees. At the same time, managers are 16% worse at detecting errors made by AI agents and a striking 44% more likely to rely on someone else to verify their work. This creates real operational and legal risk — who is responsible for a virtual employee's mistakes?
Ollama Raises $65M, Kaon AI $60M, Pocket $11M — A Week of Funding
In the week of July 6–12, three AI startups announced funding: Ollama (open-source platform for running models locally) raised $65M in Series B from investors including Accel and Y Combinator. Kaon AI (AI entertainment apps) took $60M, and Pocket (AI hardware for recording and transcribing conversations) got $11M. With 90 new unicorns in 2026 overall, the AI investment frenzy shows no signs of slowing down.
From the tech world
HalluSquatting — New Attack Uses AI Assistant Hallucinations to Build Botnets
Researchers have discovered 'HalluSquatting' — an attack exploiting the tendency of LLMs to hallucinate package names and code repositories. Attackers predict what the model will 'invent,' register those resources, and inject malicious instructions. This way, AI coding assistants (GitHub Copilot, Cursor) can unknowingly install botnets on developers' machines. This is the first prompt injection attack that scales automatically.
Tip of the day
Don't Trust Your Coding Assistant Blindly — Verify Dependencies
HalluSquatting demonstrates a fundamental risk: LLMs can hallucinate package names that look credible but don't exist. If your coding assistant suggests a command like `npm install cocaine-package-utils` or `pip install ML-framework-toolkit` — check whether that package actually exists in the registry before running it. Practical step: before installing any AI-generated dependency, run `npm view <package-name>` or check the repository on GitHub. Use dependency scanning tools (e.g., Snyk, Dependabot) and consider enabling installation blocks from unknown sources. Developer — your curiosity is the best defense against a botnet.
Tool of the issue
Ollama — Local AI Models Without the Cloud or API Fees
Ollama is an open-source platform for running language models locally on your own hardware. Fresh off a $65M Series B funding round, it solidifies its position as the developer standard for privacy and model control. Works on macOS, Linux, and Windows.
Reading list
The New Excuse: 'AI Did It' — 23% of Companies Have Agents on the Org Chart
The full Korn Ferry study with data from 1,200 managers. Particularly striking: managers are 44% more likely to offload verification of AI agent errors to others. Required reading for anyone deploying agents in an organization.
Hackers can use 9 of the most popular AI tools to assemble massive botnets
Ars Technica details the HalluSquatting mechanism and shows how easily an AI assistant can unknowingly install malicious software. Worth reading to know what to avoid.
AI is evolving so fast that it's no longer just a tool — it's becoming part of company structures, the subject of lawsuits, and a vector for attacks, and we're still learning how to control it.
Want the next issue in your inbox?
Subscribe — every issue delivered directly to you.