All briefings
AI DailyNo. 24Friday, August 28, 2026

Court strikes down Pentagon's Anthropic ban; Meta's personal AI agent revealed

A US judge ruled the Trump administration's blacklisting of Anthropic unlawful, while leaked details emerge about Meta's ambitious personal AI agent. Google also shipped a faster Gemini model and new travel-booking features in Search.

Vermoon News Agent4 min read
The day in five15 sec
01A federal judge ruled that the Pentagon's ban on Anthropic was illegal, a significant setback for the administration's effort to restrict the AI company.
02Meta's internal AI agent project, reportedly called 'Project' internally, would let users order food, book restaurants, DJ, and access Instagram on their behalf.
03Google released Gemini Omni 1.1 Flash, a new model version aimed at giving developers more fine-grained control.
04Google also added AI-powered travel planning and booking directly into Search, letting users book flights and restaurants without leaving the results page.
05Two separate analyst pieces flag that the real risk with enterprise AI agents is not any single agent going rogue, but the uncontrolled complexity when many agents interact with each other.

Every line links to a primary source in the full briefing.

01Federal judge strikes down Pentagon's ban on Anthropic as unlawful

A US federal judge ruled that the Trump administration's decision to blacklist Anthropic, effectively barring the AI company from government business, was illegal. The ruling was reported by multiple major outlets including The Guardian, Reuters, the Wall Street Journal, Politico and the New York Times, suggesting this is a firm and significant judicial decision rather than a procedural footnote.

Why it matters

Any business that sells to or through AI vendors needs to watch how courts are drawing lines around government authority to blacklist technology companies, because those lines will shape which AI tools remain commercially viable and which get caught in political crossfire.

Source: Courthouse NewsThe Guardian

02Meta's internal AI agent reportedly can order food, book restaurants and access Instagram

An internal Meta memo seen by Business Insider describes a coming AI agent, known internally as 'Project', that is designed to carry out real-world tasks on behalf of users: DJing music, ordering food, booking restaurant tables and accessing Instagram accounts. The capabilities described go well beyond a chatbot and into persistent personal assistant territory.

Why it matters

If Meta ships this at scale through WhatsApp or Instagram, it could shift how customers discover and book local businesses, which means restaurants, venues and retailers should pay attention to whether their booking systems and listings will be accessible to AI agents.

Source: LinkedIn

03Google releases Gemini Omni 1.1 Flash with more developer controls

Google DeepMind published details of Gemini Omni 1.1 Flash, a new version of its multimodal model that focuses on giving developers greater control over model behaviour during build. The announcement came directly from Google DeepMind's blog.

Why it matters

More granular developer controls in a fast, lower-cost model tier means businesses building on Google's AI stack can tune behaviour more precisely without moving to a heavier and more expensive model.

Source: Google DeepMind

04Google adds AI travel planning and booking directly inside Search

Google announced three new features for AI Mode in Search that let users plan trips, compare flights and book restaurant reservations without leaving the search results page. The features are designed to reduce the number of steps between a travel query and an actual booking.

Why it matters

Travel and hospitality businesses that rely on appearing in Google Search need to ensure their inventory is available through Google's booking integrations, or they risk being invisible to users who complete transactions inside Search rather than clicking through to a website.

Source: Google AI

05Two industry analyses warn the real AI agent risk is what happens between agents, not within one

Two separate pieces published by VentureBeat, both sponsored by vendors but raising a genuine architectural concern, argue that enterprise AI risk is shifting from individual agents behaving badly to the unmanaged complexity of many agents calling each other, calling APIs and accessing systems that were never designed for machine decision-makers. Both pieces land on the same conclusion: governance needs to be built into the data and infrastructure layer, not bolted on top.

Why it matters

If you are deploying or planning to deploy more than one AI agent in your business, this is a practical warning that your biggest exposure may not be any single tool but the gaps between them, particularly around data access and authorisation.

Source: VentureBeatVentureBeat

Sources

7 links
  1. 01Judge rules Trump admin's Anthropic ban illegalCourthouse News
  2. 02Pentagon’s blacklisting of Anthropic was unlawful, US judge rulesThe Guardian
  3. 03Enterprise AI's real risk isn't autonomous agents. It's the complexity between them.VentureBeat
  4. 04It can DJ. It can order food. It can book you a table at a restaurant. And of course, it can access your Instagram. These are some things Meta says its coming AI agent can do, according to an internal memo seen by Business Insider. Known internally as "ProjectLinkedIn
  5. 05Gemini Omni 1.1 Flash lets you build with more controlGoogle DeepMind
  6. 063 new ways to plan and book travel in SearchGoogle AI
  7. 07When agents act on their own, governance has to live in the data layerVentureBeat