Microsoft JUST BROKE OpenAI...
Credibility score: 49/100 — Mixed Credibility. Several questionable claims detected. Watch with healthy skepticism.
BSmeter analyzed "Microsoft JUST BROKE OpenAI..." and rated it 49/100 for credibility (a BS score of 51/100 — mixed credibility), on 2026-06-04. Its weakest claim — "Fine-tuned Microsoft models will be 10x cheaper than GPT-5 with proprietary moat" — scored 35/100 and was flagged as sketchy. 19 claims were checked against the video transcript. Scores are produced by BSmeter's AI analysis of the transcript, not independent human verification.
Of 19 claims analyzed: 2 scored under 40, 16 between 40 and 69, and 1 at 70 or above.
Claims analyzed
Microsoft is swinging hard with new models and an OpenClaw partnership; Bernie wants to tax AI IPOs. — Just Vibes (50/100)
At 0:00
They're throwing everything at the wall this week! — Microsoft breaking up with OpenAI *and* Bernie Sanders wanting to tax them? Wild combo. 🤯🤖💰
Why this score: High energy news dump. The core claims are solid based on context: - Microsoft/OpenAI: Web context confirms they 'broke up' (or are separating) and Microsoft is pushing new initiatives. - Partnership: OpenClaw integration into Windows is a major visible move. - Bernie Sanders: His desire to tax AI IPOs aligns with his known political stance on tech giants.
Original quote: “All right, looks like we're going to have another insane week in AI. Probably the biggest piece of news is that Microsoft is back and they just came out swinging. Tons of new AI models, a completely different approach to developing AI as well as a massive partnership with OpenClaw. OpenClaw is now…”
Scorsese using Flux AI to produce actual Hollywood films — Dubious (40/100)
At 0:30
Storyboard tool ≠ film production — wording doing heavy lifting 💀
Why this score: Conflates pre-production tool with final output. Problems:
- Speaker admits it's called "storyboarding and scene exploration"
- No evidence Flux is generating finished shots for release
- Community notes this is standard concept art use, not "producing films"
Original quote: “Martin Scorsesi is working with Flux to help him storyboard some of the movies to generate some of the shots. They're calling it storyboarding and scene exploration, but really what it is is using AI to produce Hollywood films”
Big hedge funds sold Microsoft stock over AI fears — Dubious (45/100)
At 2:30
Vague 'a lot of big hedge funds' — zero names or filings cited 💀
Why this score: No specific evidence provided. Key issues:
- Speaker names no hedge funds, no 13F filings, no dollar amounts
- Public data shows institutions remained net buyers of MSFT through 2025
- Claim rests on anonymous market chatter rather than verifiable transactions
Original quote: “Microsoft really surprised me with this one. So, Microsoft, I think, was seen by a lot of people as potentially being kind of the casualty of this AI race. A lot of the coding models probably would have put some dent into its software as a service business. Again, obviously not everyone believes…”
Microsoft's strategy of trailing the frontier is potentially killer — Opinion (50/100)
At 4:30
Bold prediction — we'll see if the 'killer' part lands 💀
Why this score: Pure opinion on future outcomes.
- Speaker frames deliberate lag as clever cost-saving
- No data offered on whether this actually beats racing to the front
- Title already calls it a 'breakup,' so the narrative is baked in
Original quote: “So, while Microsoft isn't at the top, it's hard to say that they're falling behind. But here's the thing, their playbook isn't about getting to the top. And I got to say this strategy, if it works out the way they plan, could be in a word killer.”
Microsoft's inference chip announced January 2026 — OK (60/100)
At 6:30
Timing matches recent reports but details still thin.
Why this score: Recent context shows Microsoft pushing hardware independence.
- No public January 2026 announcement found in current coverage.
- Build 2026 coverage focuses on post-OpenAI separation strategy.
Original quote: “This is more for running the models once they've completed training. Looks like this was announced in January 2026.”
Microsoft introduced Frontier Tuning for custom models — Dubious (45/100)
At 6:43
Sounds like rebranded fine-tuning with zero public proof.
Why this score: No independent verification of 'Frontier Tuning' term.
- Current Microsoft/OpenAI coverage mentions custom model work but not this name.
- Could be internal branding or future roadmap item.
Original quote: “And they also talked about something they call Frontier Tuning, which is rather interesting.”
Microsoft's model benchmarks are self-reported only — Opinion (50/100)
At 6:55
Fair skepticism — third-party testing hasn't happened yet.
Why this score: Standard industry problem — companies often release internal numbers first.
- No contradiction with known facts.
- Just noting the usual 'trust us' phase before external validation.
Original quote: “A lot of the benchmark tests for these models, they're Microsoft reported, right? So, we haven't got our hands on them.”
Anthropic faced major backlash over pirated training data — OK (65/100)
At 7:25
True-ish but timing and scale feel overstated.
Why this score: Anthropic has faced data lawsuits but 'a lot of trouble' is vague.
- Multiple AI labs have similar legal exposure.
- No unique recent scandal singled them out in current coverage.
Original quote: “Anthropic got in a lot of trouble for pirating some of the data that the models were trained on.”
Enterprises can use general 'frontier models' that are universally smart. — Solid (75/100)
At 8:30
They're all just... good? — That sounds like marketing fluff, but the premise is sound. 🤔👌
Why this score: Frontier models are inherently designed for broad capability across domains. - The speaker acknowledges that while they are 'good at everything,' this generality isn't guaranteed or defined.
Original quote: “So what this means is let's say you're a business, a large enterprise, you can take one of the frontier models that are good at everything.”
Custom models reach near-SOTA performance with 10x efficiency — Dubious (45/100)
At 10:30
Big claim, zero numbers or benchmarks shown — just vibes so far ⚠️
Why this score: No specific evidence provided.
- Speaker asserts "as good as the best" without naming models or metrics
- 10x efficiency figure appears pulled from thin air
- Web context shows Microsoft/OpenAI tensions but no public data backing this exact performance claim
Original quote: “the end result is something that's as good as the best in the world models from just a few months ago, but 10x more efficient”
Fine-tuned Microsoft models will be 10x cheaper than GPT-5 with proprietary moat — Sketchy (35/100)
At 10:45
10x cheaper sounds great — until you realize no pricing or model actually exists yet 💀
Why this score: Future promise dressed as current fact.
- "Microsoft AI model 4999" is not a real released product
- Efficiency claims lack any cited benchmarks or internal Microsoft data
- Recent reports confirm Microsoft/OpenAI relationship shift, but no evidence of 10x cost advantage materializing
Original quote: “your Microsoft AI model 4999 9 will be almost as good, 10x cheaper and the process through which you built it for that specific use case will be your mode”
Custom frontier tuning will outperform off-the-shelf models for enterprises — Opinion (50/100)
At 11:05
Classic "trust the process" take — heavy on faith, light on proof 🤷♂️
Why this score: Pure speculation presented as likely outcome.
- No comparative studies or enterprise case results cited
- Web sources discuss the breakup dynamics but don't validate custom tuning superiority
- This remains an untested prediction, not established fact
Original quote: “it's very likely that as people get proficient with this, this approach will be far far far more effective than just getting some model off the shelf”
Microsoft has a new chip called Maya 200 — Dubious (35/100)
At 12:30
No Maya 200 chip — sounds like a mix-up with existing hardware 💀
Why this score: No public record of a Microsoft chip named Maya 200.
- Microsoft’s current AI silicon is Maia 100 (announced 2023).
- No announcements or leaks mention a “Maya 200” as of June 2026.
Original quote: “Microsoft chip Maya 200 or whatever it's called”
Microsoft's Frontier Tuning lets companies train custom models on their private data — OK (55/100)
At 14:30
Sounds plausible but still marketing copy — actual results TBD
Why this score: No independent verification yet.
- Based solely on Microsoft's own blog post
- Speaker correctly notes it's "ideas on paper" with anecdotes only
- Real performance data will take time to emerge
Original quote: “This is from the Microsoft blog post about this particular thing. They're saying that in this setup that we're talking about, the most valuable data is yours. The the enterprise, the business, the trace of real work an agent completes, the sequence of steps, the decisions, the actions taken to…”
Microsoft's new model fixes enterprise control, cost, and data issues better than closed or open-source options — Opinion (50/100)
At 16:30
Classic vendor-lock-in pitch dressed up as enterprise salvation 💼🔒
Why this score: Opinion, not fact.
- Speaker presents Microsoft's offering as the clear winner without naming the actual product or showing benchmarks
- Real-world enterprise AI decisions involve far more variables than "control + cheap + private"
- No evidence presented that this beats fine-tuned open models on cost or security
Original quote: “But I got to say that for enterprise use cases, this does seem to solve a major major problem. If you have one of the frontier AI models like Claude or Chad GPT, you know what I mean? One of the GPT models, you don't get too much control over it. It's still very much closed. And yeah, I guess you…”
Maya silicon makes AI training cheaper — Dubious (45/100)
At 18:30
No public pricing or benchmarks shown yet 💸
Why this score: Cost claim lacks evidence.
- Maya silicon mentioned but no cost-per-token figures
- No comparison to competitors' hardware
- Web search shows no public Maya pricing data as of June 2026
Original quote: “It runs cheaper on Microsoft's Maya silicon on those chips that we talked about.”
Microsoft creating MXC sandbox containers for AI agents in Windows — Dubious (45/100)
At 20:30
No public evidence of 'MXC' containers yet 🚩
Why this score: Name and details unverified.
- Web search shows no Microsoft announcement or documentation for 'Microsoft execution containers' or 'MXC'
- Recent coverage focuses on Microsoft/OpenAI relationship changes, not new sandbox runtime
- Claim may be speculative or based on unannounced internal work
Original quote: “So the big kind of agentic thing that Windows is doing here is they're creating these sandbox environments to run these agents. So Windows becomes an agent runtime and Microsoft is calling these the Microsoft execution containers MXC.”
Defines autopilot AI as fully autonomous 24/7 agents — Dubious (45/100)
At 22:30
Calling background agents "24/7 autonomous" while current tools still need heavy human oversight.
Why this score: Terminology stretch.
- Current Microsoft Copilot and similar agents require prompts, approvals, and monitoring.
- No public evidence of fully hands-off 24/7 autonomous enterprise agents at scale yet.
- Distinction between co-pilot and autopilot is marketing framing, not technical reality in 2026.
Original quote: “co-pilots is something that kind of goes along with you and builds alongside with you. And autopilots is something that just runs autonomously in the background 24/7.”
Microsoft internal doc calls phase one 'make people addicted' — Unverifiable (50/100)
At 24:30
No public copy of the document — just the speaker's word.
Why this score: No independent source confirms the document exists.
- Speaker references "claw pilot overview and plan with project lobster" but provides zero link or screenshot.
- Microsoft has not commented on any such strategy doc.
- *Unverifiable* until the actual file surfaces.
Original quote: “that claw open shell whatever it's kind of like how everything evolves into crabs everything in tech evolves into claws or AI agents so this scout connects to your your teams your your one drive all of the the calendars and everything everything everything behind it and interacts with you via teams…”
See the full analysis with timestamps →