StationX
MEMBERSHIP

Sec & AI News — 27 March 2026

8 min readBy Nathan House
Get every new Sec & AI News issue
Straight to your inbox. No spam.

🟠 OpenAI Pulls the Plug on Sora, Blindsides Disney

Sora is dead. App, API, whole video business — gone overnight. Three months after a $1B Disney deal for Mickey Mouse videos, OpenAI killed it without warning. Disney walked. Teams were still collaborating Monday; Tuesday it didn't exist. Turns out $15M/day in costs against $2.1M in revenue does that.

🟣 Anthropic Ships 74 Releases in 52 Days — and Leaks "Mythos"

Anthropic is shipping like sleep is optional. 74 releases in 52 days: Computer Use drives your mouse (slowly), Dispatch texts your computer from your phone, Auto Mode stops Claude Code asking permission for harmless commands. Then it left an unprotected blog draft on its own site. Fortune found it. A new tier above Opus called Mythos — a "step change" that can "exploit vulnerabilities in ways that far outpace defenders." Rolling out cautiously. Very expensive to serve.

🏛️ Federal Judge Blocks Pentagon's "Orwellian" Anthropic Ban

A federal judge halted the Trump administration's move to brand Anthropic a supply chain risk — a label normally reserved for foreign adversaries. The trigger: Anthropic tried to stop Claude being used in autonomous weapons. The Pentagon didn't like that. Judge Rita Lin called the measures "Orwellian" and potentially "crippling." One battle won. A second designation still pending in the DC circuit.

🔵 Google Ships Gemini 3.1 Flash Live and Lyria 3 Pro

Gemini 3.1 Flash Live is the conversational one — webcam, screen share, voice — now in the API, Enterprise, Search, Gemini Live, and AI Studio. Google also demoed a vibe-coded browser that builds whole web pages live from any query. Impressive. Useless. No memory, no persistence. Meanwhile Lyria 3 Pro grew up: 3-minute tracks with intros, verses, choruses, and bridges, up from 30 seconds. Not replacing producers yet.

🎙️ The Voice Wars Heat Up: Suno, Smallest.ai, Mistral

Everyone wants your voice. Suno 5.5 lets Pro and Premier users train their own voice into the model — record a sample, verify, generate songs with your tone and vibrato. Smallest.ai's Lightning V3 is a TTS model for voice agents that handles pauses and disfluencies, and beats ElevenLabs on most metrics — the benchmark they made. Mistral's Voxtral TTS is open-weights: clone a voice from 3 seconds of audio and run it locally for free.

🛒 OpenAI Builds Shopping Into ChatGPT

OpenAI wired product discovery into ChatGPT via its Agentic Commerce Protocol — visual shopping, comparisons, merchant feeds. Businesses list products free. For now. Their ad platform couldn't prove ROI to early advertisers, so this looks like the groundwork for sponsored placements. Google Shopping for chatbots.

🎨 Image and Video Tools: Lovart Moves Objects, Seedance Skips the US

Lovart AI shipped a move-object feature — grab an object, drag it, the model regenerates the scene. Pair it with a video generator for start-to-end frame animation. Free tier available. Over in CapCut, ByteDance's Dreamina Seedance 2.0 landed with safety limits (no real faces) — but not in the US or Europe, courtesy of Hollywood's copyright complaints. Reports say it's impressive. Most of us can't verify.

📚 Wikipedia Bans AI-Generated Articles

Wikipedia editors voted to ban LLMs from writing or rewriting articles. Two exceptions: copy editing and translation, both needing human review. When models trained on Wikipedia start writing Wikipedia, you get model collapse. The encyclopedia saw it coming.

🤖 Figure 03 Robot Walks Into the White House

First humanoid robot at the White House. Figure AI's Figure 03 walked alongside First Lady Melania Trump at the Fostering the Future summit. The $25,000 robot waved, said some things, and left. Nobody was harmed.