Sent to 16 readers
Race to AGI Daily Digest - Wednesday, December 24, 2025
Share:
TLDR
Grok goes operational on GenAI.mil, giving 3 million Pentagon users direct access to xAI’s model on a secure domain.
OpenAI frames prompt injection as a long-term structural risk for AI browsers and agents, reinforcing the need for built-in defenses.
Taiwan’s AI Basic Act sets a national AI strategy and safety rules, adding another pole to an increasingly fragmented global AI regime.
OpenAI, Google, Meta, Anthropic, xAI, and Perplexity face a lawsuit over ‘pirated books’ training data, escalating the copyright fight.
The Full Story
Following Monday’s Mango-and-Avocado reveal and Tuesday’s world-model push from AMI Labs, today is about where these systems actually live: in militaries, laws, and court filings.
On the military side, xAI’s Grok has been switched on for 3 million Pentagon users via GenAI.mil, a government-only AI portal. This is one of the first times a frontier-style model is wired directly into day-to-day defense workflows. It’s not a demo anymore; it’s baked into how analysts and staff get answers. More on the Grok–Pentagon rollout ->
That makes Tuesday’s security storyline feel sharper. OpenAI is now calling prompt injection a long-term threat for AI browsers and agents, basically saying: once models roam the open web for you, the web can talk back and hijack them. It’s a clear continuation of the “AI security as a first-class product” arc we’ve been tracking. OpenAI’s prompt-injection warning ->
Governments are starting to codify all this. Taiwan’s new AI Basic Act lays out a national strategy and safety rules, adding another node to the “Global AI Infrastructure Fragmentation Intensifies” narrative, where each region wants its own chips, clouds, and laws. Taiwan’s AI Basic Act summary ->
In the courts, OpenAI, Google, Meta, Anthropic, xAI, and Perplexity now share a different kind of battlefield: a lawsuit over training on “pirated” books. That’s the flip side of Monday and Tuesday’s model-wars story—whoever wins on capabilities still has to answer for the data diet that got them there. Details on the ‘pirated books’ lawsuit ->
Markets keep rewarding the rails. NVIDIA, Broadcom, and Amazon are all up, even as C3.ai and Oracle slip, which lines up with our earlier view that investors prefer chips and cloud over pure-play AI names. Check NVIDIA’s full AI profile ->
And while all this plays out at the top, HOPE AI’s partnership with 100 Black Men is a reminder that AI literacy is becoming civic infrastructure: if models write more of what we see, people need the tools to judge it. HOPE AI’s literacy push ->
Get This Delivered Daily
Join thousands of AI professionals who start their day with Race to AGI.