Anthropic just ran the cleanest âhave your cake and donât ship itâ play weâve seen in AI.
Iâm looking at the Mythos + Opus 4.7 rollout and all I see is one thing:
We just watched a company ship a BANGER model⌠and still get the press to obsess over the one they didnât release.
Bloomberg running âToo Dangerous For Release.â Axios going with âconcedes it trails unreleased Mythos.â CNBC: âless risky than Mythos.â
This is exactly the narrative @AnthropicAI wanted. And they got it.
1. Genuine Capability â Do we buy the zero-day flex?
Letâs start with the red.anthropic.com Mythos preview: thousands of zero-days in every major OS and browser; capable of hacking banks if misused.
Weâre all thinking the same thing: this is either the most cracked red-team engine ever built⌠or the most overclocked marketing copy since âself-driving is basically solved.â
Hereâs where I land:
- We know Opus 4.7 is real and nasty. Beats GPTâ5.4 and Gemini 3.1 Pro on agentic coding, scaled tool use, agentic computer use, and financial analysis. Thatâs not fluff; thatâs empirical.
- We know Anthropic claims Mythos is strictly better than Opus 4.7. Internal frontier model, stronger at security tasks.
- We know thereâs enough heat that the Trump administration (!!) got briefed pre-launch. Thatâs not your standard âwe found a prompt injectionâ situation.
So do I 100% verify the âthousands of zero-daysâ claim? No. Itâs coming from Anthropic itself. No third-party vuln DBs or independent security orgs have dropped the receipts.
But do I think theyâre bluffing? Also no.
Why? Because if you lie that big in security, someone eventually calls you on it. And because we already see LLMs creeping into offensive sec â open tools are already finding soft vulns at scale. Itâs not a stretch that a frontier, non-safetied internal model plus tools can sweep codebases and fuzz attack surfaces way beyond human bandwidth.
Genuine Capability â Mythos claims:
- Believability: 8/10
- Independently Verified: 3/10
- Net Capability Rating: 7.5/10
Basically: I think the core âthis thing is dangerously good at offenseâ is true. The exact numbers are probably marketing-smoothed. But itâs not sciâfi.
2. Safety Posture â Project Glasswing: shield or stage prop?
This is where everyone starts posturing.
Anthropic: âMythos is too dangerous for public release. Weâre routing it through Project Glasswing to critical infra partners and defenders first.â
Bloomberg amplifies the âToo Dangerousâ frame. Scientific American and SFist are like âwhy are experts worried?â CNBC echoes âless risky than Mythosâ for Opus 4.7. Safety theater catnip.
Hereâs the real question: is Glasswing actually constraining risk, or just a clean PR container?
On the âthis might be real defenseâ side:
- They really did not ship Mythos broadly. Thatâs a non-trivial revenue hit. When everyone else is racing to jam frontier models into every SaaS flow, holding back is costly. I respect that.
- They built the first automated system to detect and block prohibited cyber requests in Opus 4.7. Thatâs actual engineering, not just vibes. Plus the Cyber Verification Program for vetted pentesters â a real concession to the builder side.
- Glasswing going to critical infra + select open-source defenders is the right directionally correct move. You want your best offensive model in the hands of people who can harden, not random Discord kids trying to âpwn Chase for memes.â
On the âthis might be theaterâ side:
- âWe gave our ultra-offensive model to âcritical industry partnersââ is doing a LOT of narrative work. Who exactly? What guardrails? What audit? Whole thing is opaque.
- Thereâs a tension: you canât brag âthis can hack major banking systemsâ and then claim your governance is airtight when we have zero transparency on access logs, oversight, escalation paths, etc.
- Briefing the administration makes it sound serious, but letâs be honest: governments love scary frontier AI stories. It justifies budgets and controls. Thatâs theater-adjacent by default.
My read: Glasswing is partly real safety and partly PR container. The cap table gets to tell regulators, âSee, weâre the adults,â while still running the crazy stuff internally and in restricted channels.
Safety Posture â Project Glasswing:
- Actual Risk Reduction: 7/10
- PR / Optics Value: 9/10
- Theater Factor: 6/10
- Net Safety Rating: 7/10
Itâs not fake. But it is curated. And we shouldnât pretend itâs some bulletproof âwe solved misuseâ framework.
3. Marketing Genius â âWe have something better and you canât have itâ
This is where you have to tip the cap.
Most companies ship their best model and hope the press notices. Anthropic shipped Opus 4.7 â which, again, beats GPTâ5.4 and Gemini 3.1 Pro on real benchmarks â and still got every major outlet to define it relative to a product that doesnât exist publicly.
Bloomberg: âToo Dangerous For Release.â Axios: âconcedes it trails unreleased Mythos.â CNBC: âless risky than Mythos.â
That is textbook frame control.
They turned Opus 4.7 â already S-tier â into the âsafe daily driverâ and Mythos into the dragon in the basement. You know who does that? Console makers with dev kits. Weapons labs. Luxury brands with invite-only lines.
They also pulled the craziest move of all: kept Opus 4.7 pricing flat at $5 / $25 per million tokens while heavily implying âweâre holding back a supercar.â So devs get a better model than GPTâ5.4/Gemini 3.1 Pro on agents and money stuff, no price hike, and the whole world walks away thinking Anthropic is sitting on a nuke.
Legit counterargument: this is flirting with fear-mongering. The âtoo dangerous for releaseâ narrative can and will be used to justify regulation and gatekeeping that lock out open-source and small shops. And Anthropic is not sad about that.
But strictly on marketing craft?
- Narrative Control: 10/10
- Brand Positioning (responsible but frontier): 9/10
- Dev Appeal (Opus 4.7 features + pricing): 8/10
- Marketing Genius Rating: 9.5/10
They got the âtoo powerfulâ halo and the âsafer than Mythosâ trust badge in the same cycle. Thatâs not an accident. Thatâs a coordinated run with press, policy, and product all aligned.
4. Competitive Pressure â Does this corner OpenAI and Google?
This is the spicy one.
By openly saying âOpus 4.7 trails Mythos,â Anthropic basically told the world: we have internal models beyond what we sell you, and weâre responsible about not shipping them.
Axios literally framed it that way. CNBC calls Opus 4.7 âless risky than Mythos.â The implicit question lands squarely on @sama, @OpenAI, @GoogleDeepMind:
âDo you also have Mythos-class internal models youâre not talking about? If yes, why arenât you being this transparent? If no, are you behind Anthropic on both capability and safety?â
Thatâs the trap.
If OpenAI or Google admits they have stronger internal models, they invite the same âtoo dangerous for publicâ scrutiny and regulator attention. Now youâre on the back foot explaining safety posture to the same reporters Anthropic just briefed.
If they donât admit it, Anthropic gets to wear the âweâre the cautious grownups at the frontierâ crown â especially with things like the cyber-blocking system in Opus 4.7 and the Cyber Verification Program. Feeds perfectly into the âConstitutional AI, safety-firstâ mythology theyâve been building since day one.
Realistically, OpenAI and Google also have scary internal stuff. Agents, long-horizon tool use, internal security experiments. Weâre not children. But theyâve been quieter on âthis is too hot to shipâ specifics.
Mythos forces a choice:
- Either admit youâre doing similar and walk into the safety spotlight,
- or keep quiet and let Anthropic own the âsafest frontier playerâ narrative in D.C. and Brussels.
Competitive Pressure:
- On Capability Perception: 8/10 (Mythos halo makes Anthropic look like itâs leading edge)
- On Safety Narrative: 9/10 (everyone else now looks less transparent)
- On Actual Product Adoption (Opus 4.7 today): 7/10
- Net Competitive Pressure Rating: 8.5/10
Very fair counterpoint: this can backfire. If regulators
Stay sharp. â Max Signal