AI Tools & Prompts EN

Daily Brief — AI tools, product updates, and legal flash — 2026-03-22

Updated: 2026-03-22 (UTC)

Top headlines

  • Gemini’s new task automation is impressive but limited — hands-on testing on Pixel 10 Pro and Galaxy S26 Ultra shows the assistant can use apps for you, though the feature is slow and constrained to a small set of actions. (The Verge)
  • Hachette pulled the horror novel “Shy Girl” citing concerns the text was generated by AI, underscoring rising publisher caution around AI-authored works. (TechCrunch)
  • An anonymous Substack accuses compliance startup Delve of persuading customers they were “compliant” when they were not, raising questions about vendor claims and auditability. (TechCrunch)
  • New court filings show Anthropic argued it and the Pentagon were nearly aligned on safety — a counter to public statements that the relationship was irreparably broken. (TechCrunch)
  • Legal dispute: a Halide co-founder is suing a former partner who joined Apple, alleging the bringing of source code after the hire. (The Verge)
  • Opinion and ethics: a feature on generative video model Sora raises deep concerns about how some generative AI narratives intersect with dangerous ideology. (The Verge)
  • Context: it’s the 20th anniversary of the first tweet, a reminder of how platforms and public discourse evolve alongside tech. (TechCrunch)

Product & tools notes

  • Gemini task automation: useful for hands-off flows (orders, ride apps) but test for latency, error handling, and permission scopes before trusting it in production.
  • Device AI: Apple continues to bake AI into audio hardware (AirPods features reported elsewhere this week), reinforcing that edge-device AI is moving from novelty to everyday UX.

Policy, legal & ethics

  • Publishing: Hachette’s pull illustrates increasing publisher liability and reputational risk around AI-generated content; expect stricter provenance and disclosure demands.
  • Vendor claims: the Delve accusation is a cautionary tale — compliance assertions must be provable with logs, evidence, and third-party audits.
  • Employment & IP: the Halide lawsuit highlights risks when engineers move between companies with access to codebases; maintain clear IP agreements and exit checklists.
  • National security & models: Anthropic’s filings show adversarial scrutiny between AI firms and governments continues to shape access and procurement.

Practical workflows (for product and engineering teams)

  • Require provenance metadata and human-in-the-loop review for any externally-published content claimed to be authored or assisted by models.
  • Add vendor due-diligence steps: request audit artifacts, run independent compliance checks, and include contractual remedies for false claims.
  • For UI automation features (e.g., Gemini): start with scoped pilot projects, capture full interaction logs, and build robust rollback/error-handling paths.
  • For hires with sensitive access: enforce code-transfer checks, documented handoffs, and explicit IP assignment records during offboarding/onboarding.

Key takeaways

  • Generative AI is moving from demo to production but remains brittle — test extensively and log everything.
  • Claims of compliance or authorship must be demonstrable; ambiguous vendor statements or unlabeled AI content are now high-risk.
  • Legal and policy pressures are accelerating: firms, publishers, and governments are all tightening scrutiny.

Sources

Not financial/professional advice

Sources