AI Weekly Briefing: Mouse Ears and Model Wars

In this week’s 10 minute AI news roundup, I break down a headline grabbing Disney move that could redefine what “legal generative AI” looks like at scale. Disney is reportedly investing $1 billion in OpenAI, negotiating warrants for optionality, licensing a vault of iconic characters for Sora, and rolling out ChatGPT across the company with strict guardrails. But the backlash is already here, with the Writers Guild of America warning this is a turning point in the fight over creative labour and value.

Then we jump into the model wars. OpenAI ships GPT 5.2 with a massive 400k context window and tiered variants for speed, deep analysis, and maximum accuracy, while Google counters with Gemini Deep Research and an Interactions API designed to embed autonomous research workflows into products. We also cover a major UK partnership with Google DeepMind aimed at accelerating science through automation, plus why Microsoft’s new Copilot pricing for small businesses, Nvidia’s move to buy the company behind Slurm, and a fresh wave of copyright lawsuits all point to one thing: the legal and infrastructure layers of AI are now product critical.

Episode Transcript

(00:00):
Hi everyone.

(00:00):
I'm Andy and this is the AI Breakdown.
Welcome to your weekly news edition where I'll cover what happened in AI last week, why it matters all in about 10 minutes. 4 00:00:14,334.959727273 --> 00:00:15,564.959727273 Alright, first story. 5 00:00:15,924.959727273 --> 00:00:21,444.959727273 Disney just did the most Disney thing imaginable, and I mean that as both a compliment and a warning sign. 6 00:00:21,834.959727273 --> 00:00:27,834.959727273 They're invest in $1 billion in OpenAI, and they've also negotiated warrants to purchase additional equity. 7 00:00:27,894.959727273 --> 00:00:32,154.959727273 If those warrants are crucial, there are essentially options to increase their stake. 8 00:00:32,364.959727273 --> 00:00:37,704.959727273 If open AI's valuation keeps climbing, Disney is buying optionality, not just equity. 9 00:00:38,214.959727273 --> 00:00:42,924.959727273 And on top of this, as part of a three year deal, they're licensing hundreds of characters. 10 00:00:43,284.959727273 --> 00:00:46,499.959727273 Star Wars, Marvel, Pixar, the whole treasure chest. 11 00:00:47,64.959727273 --> 00:00:55,134.959727273 So open AI's image and video generator, Sora can create content featuring icons like Mickey Mouse and Cinderella starting early next year. 12 00:00:55,494.959727273 --> 00:00:59,394.959727273 And importantly, it excludes real actors, likenesses or voices. 13 00:00:59,904.959727273 --> 00:01:02,604.959727273 Disney's also rolling out chat, GPT enterprise wide. 14 00:01:02,994.959727273 --> 00:01:12,984.959727273 Bob Iger framed it as a way to thoughtfully and responsibly extend the reach of their storytelling through generative AI while respecting and protecting creators and their works. 15 00:01:13,524.959727273 --> 00:01:19,164.959727273 This is what legal generative AI at scale looks like when the IP is locked down and the brand is precious. 16 00:01:19,614.959727273 --> 00:01:25,404.959727273 Disney even negotiated guardrails, so characters can't be shown in inappropriate ways, but the mood isn't. 17 00:01:25,404.959727273 --> 00:01:38,154.959727273 All fireworks and mouse ears, the Writer's Guild of America came out swinging, seeing the deal appears to sanction theft of their work and seeds the value of what they create to a tech company that has built its business off their backs. 18 00:01:38,664.959727273 --> 00:01:41,64.95972727 That tension innovation versus labor rights. 19 00:01:41,694.95972727 --> 00:01:44,424.95972727 It's going to be a recurring theme in 2026. 20 00:01:45,636.77790909 --> 00:01:50,916.77790909 Next up and as expected, OpenAI launch their new model, GPT 5.2. 21 00:01:51,426.77790909 --> 00:02:07,121.77790909 In response to Sam Altman's code red declaration aimed at focusing all efforts on boosting chat, GPT speed, reliability, and personalization to counter intense competition, primarily from Google's Gemini three model, which had surpassed open AI on benchmarks. 22 00:02:08,136.77790909 --> 00:02:09,306.77790909 GPT 5.2 23 00:02:09,306.77790909 --> 00:02:15,6.77790909 is available to chat GPT plus users and via API, and it comes in three flavors. 24 00:02:15,426.77790909 --> 00:02:21,936.77790909 Instant for speed thinking for long documents and analysis, and pro for maximum accuracy on hard problems. 25 00:02:22,356.77790909 --> 00:02:32,526.77790909 The headline feature is that massive 400,000 token context window, meaning it can ingest very large documents or long multi-step instructions without falling apart. 26 00:02:33,276.77790909 --> 00:02:35,496.77790909 Open AI's product Chief Fiji CMO said. 27 00:02:35,871.77790909 --> 00:02:37,371.77790909 We designed 5.2 28 00:02:37,371.77790909 --> 00:02:48,621.77790909 to unlock even more economic value for people, and highlighted that it's better at creating spreadsheets, building presentations, writing code, perceiving images, and understanding long context. 29 00:02:48,981.77790909 --> 00:02:51,406.77790909 And the industry benchmarks are backing this up too on the whole. 30 00:02:53,109.50518182 --> 00:03:03,549.50518182 Meanwhile, Google wasn't just shipping a model, it launched Gemini Deep Research, a reimagined research agent built on Gemini three Pro, plus an interactions API. 31 00:03:03,549.50518182 --> 00:03:12,579.50518182 So developers can embed that autonomous research workflow into their own apps on one of the harshest tests to put models like this through its paces. 32 00:03:12,699.50518182 --> 00:03:17,229.50518182 Human's Last exam, the results for the latest models show a real step up. 33 00:03:17,784.50518182 --> 00:03:22,974.50518182 Gemini three deep think comes in around 41% while GPT 5.2 34 00:03:22,974.50518182 --> 00:03:26,664.50518182 thinking is miles behind at roughly 34.5% 35 00:03:26,994.50518182 --> 00:03:27,864.50518182 with others lower. 36 00:03:28,524.50518182 --> 00:03:33,774.50518182 Whilst this means that AI has not solved intelligence yet, these systems still miss more than they hit. 37 00:03:34,164.50518182 --> 00:03:36,924.50518182 It indicates agents are becoming genuinely useful. 38 00:03:37,254.50518182 --> 00:03:44,934.50518182 They're increasingly capable of doing the messy multi-step work of finding relevant info, weighing it, and stitching it into a coherent answer. 39 00:03:45,519.50518182 --> 00:03:51,189.50518182 Instead of just guessing confidently, why should you care? Well, it means faster research loops. 40 00:03:51,369.50518182 --> 00:03:59,199.50518182 You can hand off the hard work, get in the eye to find the best sources, compare them, and summarize and get a solid first pass in minutes. 41 00:03:59,619.50518182 --> 00:04:09,9.50518182 There's better multi-step help planning, troubleshooting analysis, and decision support benefit because the model is more likely to stay on the rails across several steps. 42 00:04:09,654.50518182 --> 00:04:11,154.50518182 Still not autopilot. 43 00:04:11,484.50518182 --> 00:04:17,754.50518182 A 41% score on a brutal benchmark is a reminder to treat outputs like a Sharp Junior Analyst. 44 00:04:18,174.50518182 --> 00:04:24,594.50518182 Great at drafting and digging, but you still want spot checks, sources, and same constraints for anything important. 45 00:04:26,241.77790909 --> 00:04:30,861.77790909 Now, a big UK story that's easy to miss if you are glued to the model wars. 46 00:04:31,281.77790909 --> 00:04:34,701.77790909 The UK has struck a major partnership with Google Deep Mind. 47 00:04:35,76.77790909 --> 00:04:45,696.77790909 The government says DeepMind will open its first fully automated AI research lab in the UK in 2026 using robotics and AI to accelerate scientific breakthroughs. 48 00:04:46,236.77790909 --> 00:04:55,836.77790909 The headline mission is developing next generation superconductor materials, the kind that can carry electricity with zero resistance, which could be massive energy in computing. 49 00:04:56,556.77790909 --> 00:05:09,696.77790909 The partnership also calls out Fusion Energy Research plus public service automation through a Gemini for government system to handle routine admin and education work on an AI grounded in the national curriculum to support teachers. 50 00:05:10,356.77790909 --> 00:05:20,286.77790909 Prime Minister k Starer pitched it as a way to unlock cleaner energy, smarter public services, and new opportunities, which will benefit communities up and down the country. 51 00:05:20,796.77790909 --> 00:05:22,986.77790909 Partnerships like this will help us go further faster. 52 00:05:24,21.77790909 --> 00:05:25,221.77790909 That's the sales pitch. 53 00:05:25,581.77790909 --> 00:05:30,981.77790909 The reality will be whether it actually ships working systems, not just glossy announcements. 54 00:05:31,551.77790909 --> 00:05:34,856.77790909 The public reaction, unsurprisingly, is cautiously optimistic. 55 00:05:35,631.77790909 --> 00:05:43,191.77790909 There's pride in DeepMind's UK roots, but also a bit of side eye because we've all seen big Tech plus government projects go sideways. 56 00:05:43,521.77790909 --> 00:05:57,351.77790909 The encouraging part is the deal explicitly expands work with the UK's AI Safety Institute, which at least suggests someone is thinking about safety and governance while they chase breakthroughs back to Disney for a second. 57 00:05:57,381.77790909 --> 00:06:01,581.77790909 Because the more I think about it, the more it feels like a turning point for content creation. 58 00:06:02,196.77790909 --> 00:06:02,946.77790909 Reuters noted. 59 00:06:02,946.77790909 --> 00:06:10,236.77790909 Disney sees this as a way to let fans and creators co-create short form videos, some of which might even stream on Disney Plus. 60 00:06:10,536.77790909 --> 00:06:11,346.77790909 That's wild. 61 00:06:11,706.77790909 --> 00:06:20,976.77790909 It's basically official fan fiction, but with studio grade ip, from a workflow perspective, this could change how marketing teams operate. 62 00:06:21,381.77790909 --> 00:06:32,421.77790909 Imagine ad agencies generating endless variations of short clips with licensed Disney characters legitimately, rather than dancing round copyright and hoping nobody notices. 63 00:06:33,81.77790909 --> 00:06:38,811.77790909 And internally, Disney expects this to boost efficiency, including in things like VFX and editing. 64 00:06:39,606.77790909 --> 00:06:42,606.77790909 But the labor question is still sitting there staring at us. 65 00:06:42,996.77790909 --> 00:06:49,146.77790909 The wgas reaction wasn't subtle, and it's a reminder that AI and media isn't just a tech upgrade. 66 00:06:49,506.77790909 --> 00:06:53,166.77790909 It's a negotiation about who gets paid and who gets replaced. 67 00:06:54,525.86881818 --> 00:06:56,655.86881818 Quick one that matters a lot for smaller teams. 68 00:06:56,925.86881818 --> 00:07:05,210.86881818 Microsoft launched Microsoft 365 copilot business on December 1st, aimed at organizations with fewer than 300 users. 69 00:07:05,925.86881818 --> 00:07:08,25.86881818 It's $21 per user per month. 70 00:07:08,415.86881818 --> 00:07:10,725.86881818 Cheaper than the $30 enterprise price. 71 00:07:11,25.86881818 --> 00:07:26,895.86881818 And Microsoft says it's the same feature set, so you get copilot inside Word, Excel, PowerPoint Outlook teams plus copilot chat, and tools for building custom agents all integrated with your Microsoft 3, 6, 5 data and security settings. 72 00:07:27,405.86881818 --> 00:07:34,995.86881818 There's also promotional pricing through March 31st, 2026 that can bring it down to about $18 per user via partners. 73 00:07:35,475.86881818 --> 00:07:39,765.86881818 Why does this matter? Because it drags AI adoption into the long tail of the economy. 74 00:07:40,275.86881818 --> 00:07:51,945.86881818 If you're a 20 person agency, a 50 person SAS shop, or a scrappy professional services firm, you can now standardize on an enterprise grade AI assistant without pretending you're a Fortune 500. 75 00:07:52,395.86881818 --> 00:08:04,40.86881818 Microsoft even cited research suggesting copilot could deliver return on investment of 132% to 353% for small and medium sized businesses through time savings. 76 00:08:05,70.86881818 --> 00:08:09,660.86881818 I always take vendor numbers like this with a pinch of salt, but directionally cheaper. 77 00:08:09,660.86881818 --> 00:08:14,400.86881818 Bundled AI inside the tools people already live in is how this stuff becomes normal. 78 00:08:16,138.14154545 --> 00:08:17,848.14154545 Two more quick hitters before we wrap. 79 00:08:18,178.14154545 --> 00:08:22,348.14154545 First, Nvidia bought shed MD the primary developer of Slum. 80 00:08:22,678.14154545 --> 00:08:28,828.14154545 The open source workload manager used to schedule jobs on a huge chunk of the world's biggest supercomputers. 81 00:08:29,488.14154545 --> 00:08:33,538.14154545 Nvidia says Slim runs on over half of the top 100 supercomputers. 82 00:08:33,943.14154545 --> 00:08:37,303.14154545 NVIDIA's promising it'll stay open source and vendor neutral. 83 00:08:37,513.14154545 --> 00:08:43,393.14154545 And Sheddy MD's, CEO, Danny Obel called it the ultimate validation of slums Critical role. 84 00:08:44,113.14154545 --> 00:08:52,183.14154545 This matters because scheduling is the boring software layer that decides whether your expensive GPUs are actually doing work or sitting idle. 85 00:08:53,143.14154545 --> 00:08:58,333.14154545 Nvidia owning that traffic control layer makes its grip on the AI stack even tighter. 86 00:08:58,753.14154545 --> 00:08:59,143.14154545 Great. 87 00:08:59,143.14154545 --> 00:09:01,393.14154545 If you want a smooth, integrated experience. 88 00:09:01,723.14154545 --> 00:09:04,393.14154545 Less great if you're worried about ecosystem lock-in. 89 00:09:06,46.77790909 --> 00:09:08,776.77790909 And finally, the copyright fights keep escalating. 90 00:09:09,286.77790909 --> 00:09:19,576.77790909 Disney issued a cease and desist to Google over AI use of its characters accusing the tech giant of copying its copyrighted material on a massive scale to build and deploy AI systems. 91 00:09:20,866.77790909 --> 00:09:31,756.77790909 And the Chicago Tribune sued perplexity alleging its Scrapes Tribune articles even behind paywalls and reproduces them verbatim or in detailed summaries and even claims. 92 00:09:31,756.77790909 --> 00:09:36,706.77790909 Perplexity, comet browser bypasses, paywall protections, perplexity. 93 00:09:36,706.77790909 --> 00:09:40,966.77790909 Jesse Dwyer, head of communications pushed back with a line that made me laugh. 94 00:09:41,386.77790909 --> 00:09:49,846.77790909 Publishers have been suing new tech companies for a hundred years, starting with radio, tv, the internet, social media, and now ai. 95 00:09:50,296.77790909 --> 00:09:55,906.77790909 Fortunately, it's never worked, or we'd all be talking about this by telegraph regardless of who wins. 96 00:09:56,116.77790909 --> 00:10:01,516.77790909 If you're building anything that summarizes web content or uses retrieval, this is your reminder. 97 00:10:01,576.77790909 --> 00:10:03,856.77790909 The legal layer is now a product requirement. 98 00:10:05,390.41427273 --> 00:10:07,280.41427273 That's all for this week's EI roundup. 99 00:10:07,610.41427273 --> 00:10:11,270.41427273 If you found value in this breakdown, please leave a rating and hit subscribe. 100 00:10:11,510.41427273 --> 00:10:12,305.41427273 See you next week.