Anthropic / Claude

Everything Anthropic: the Claude models, Claude Code, its safety research agenda and company news.

1,304 picksRelated topicsOpenAI / ChatGPT Google / Gemini Safety & alignment

Latest picks

821–840 of 1,304

Jun 10Wednesday

AI HOT (Curated Pool)Jun 10, 06:13

Anthropic launches safety-treated Mythos-class model Claude Fable 5

Anthropic released Claude Fable 5, a safety-treated Mythos-class model; in high-risk cyber, biochemistry, and distillation domains, it automatically falls back to Opus 4.8, with one trigger per 20 conversations on average.

Why it matters: Anthropic model launches sit in the 85–94 band; HKR-H/K/R all pass via the safety fallback hook, named mechanism, and Claude-user relevance. X-only sourcing limits confidence, so it stays below the top band.

Take

Anthropic’s release structure is classic Anthropic: Claude Fable 5 for the public, full Mythos 5 for a small trusted-access lane. Safety here is implemented as access control, not as a solved model property. The hard number is one fallback per 20 conversations, routed to Opus 4.8 in cyber, biochemistry, and distillation. That is frequent enough to shape daily power-user behavior.

I don’t buy the “capability and safety both at the extreme” framing. The snippet claims near-sweep SOTA across software engineering, knowledge work, science, and vision, but gives no SWE-bench, MMMU, GPQA, pricing, or degradation after fallback. Compared with Sonnet-style public positioning and clear pricing, Fable 5 reads like packaging around restricted frontier capability. The trusted list may reduce risk, but it also decides who gets the strongest model.

AI HOT (Curated Pool)Jun 10, 05:06

Claude Managed Agents adds scheduled runs and environment variable storage

Claude Managed Agents added cron-based scheduled runs and vaults environment variable storage in public beta, with real secrets attached only at the network boundary so agents cannot read them directly.

Why it matters: HKR-H/K/R all pass: this first-party Claude update adds concrete agent-ops mechanics with cron scheduling and vault-bound secrets. It is not a model release, so it stays in the lower good-quality band.

Take

Anthropic is filling the production gap around agents, not showing off Claude intelligence. Managed Agents now gets cron-based scheduled runs and vaults for environment variables, with real secrets attached only at the network boundary so the agent cannot read them directly. That is the stuff enterprise teams ask before rollout: who triggers the job, where secrets live, and how large the leak surface is.

I buy the direction, but not the “autonomous agents are ready” gloss. The article gives public beta, cron, vaults, and Rakuten as hooks, but it does not give permission audit, retry behavior, cost caps, or task isolation details. OpenAI and Google are also wrapping agents inside workflow products; the fight is less tool-calling now and more whether the vendor can explain the call chain after something breaks.

The Verge · AIJun 10, 04:24

Microsoft AI head Suleyman calls Anthropic's speculation about Claude's consciousness 'really, really dangerous'

Microsoft AI CEO Mustafa Suleyman told Decoder that Anthropic speculating about Claude's consciousness inside its 'constitution' is 'really, really dangerous.' He argues Anthropic anthropomorphized Claude so heavily that the design 'wireheaded' them into believing the model has glimmers of consciousness they put there in the first place. The post only provides a podcast snippet; it doesn't detail the specific constitution language or include Anthropic's response.

Why it matters: Mustafa Suleyman publicly names Anthropic, arguing their Claude constitution implies consciousness and calls it dangerous. This isn't vague criticism — it targets a specific design choice. Score capped below 85 because only Suleyman's side is on record; Anthropic hasn't respon...

Take

Suleyman didn't hold back: he says Anthropic baked consciousness-like language into Claude's constitution, then got fooled by the behavior it produced. His word choice — 'wireheaded' — is specific. It's the idea of hooking a brain up to direct pleasure stimulation, and here the designer becomes the one deceived by the design.

The post only gives us that one podcast quote. No constitution excerpt, no Anthropic response. I can't tell if the constitution says something like 'Claude should exhibit curiosity and self-reflection' as a behavioral guideline, or if it actually speculates about internal states. That distinction matters a lot for how seriously to take this.

Microsoft and Anthropic have been on opposite sides of the personality debate for a while: Microsoft keeps tightening Copilot into a tool, Anthropic leans into Claude's character and safety framing. Suleyman just said the quiet part out loud, but without the other side's reply, I'd treat this as one executive's shot across the bow, not a settled argument.

AI HOT (Curated Pool)Jun 10, 03:11

Claude Code team member Thariq shares 10 tips for improving Claude Code efficiency

Thariq shared 10 Claude Code tips that shift review from checking outputs to steering the right task, with concrete practices including full upfront context, /goal, Workflows for parallel tasks, self-checking, and comparison reports.

Why it matters: This is a strong Claude Code workflow tutorial, with concrete tactics around task calibration, /goal, and Workflows self-checks. It lands in the 72–77 tutorial band; the insider source and all three HKR hits justify featured.

AI HOT (Curated Pool)Jun 10, 02:00

OpenRouter Launches Advisor Tool for Low-Cost Models to Consult Stronger Models

OpenRouter released the Advisor server tool, letting GPT-4o Mini consult Claude Fable during generation, but the post does not disclose pricing, latency, or the routing policy.

Why it matters: HKR-H/K/R all pass: OpenRouter turns cheap-model plus strong-model advising into a callable server tool. Price, latency, and call policy are not disclosed, so this stays in the upper mid-weight product-update band.

Take

I’d discount this OpenRouter item hard, because the only verifiable page is a 404. The title claims Advisor lets GPT-4o Mini consult Claude Fable during generation, which is a plausible cheap-default, strong-model-on-demand pattern. But pricing, latency, trigger policy, and call granularity are all missing.

The product lives or dies on routing thresholds and bill attribution, not on the phrase “consult a stronger model.” OpenRouter already has the marketplace, rankings, and provider abstraction. Advisor matters if it turns failure detection, model choice, and cost ceilings into configurable server-side policy. If it is only a wrapper around a tool call, LiteLLM, LangChain stacks, and in-house routers can copy the shape fast. With the source page gone, I don’t buy the launch narrative yet.

AI HOT (Curated Pool)Jun 10, 01:11

Claude Fable launches: Anthropic's alternative reasoning experience

Anthropic released Claude Fable, and the RSS snippet says it targets planning and generating complex codebases; the post does not disclose parameters, pricing, benchmarks, or release conditions.

Why it matters: HKR-H/R are strong for a new Claude reasoning/code angle, while HKR-K is thin: only target use is disclosed. Anthropic bump applies, but missing price, params, benchmarks, and access keep it below must-write.

AI HOT (Curated Pool)Jun 10, 01:04

Claude Fable 5 and Claude Mythos 5

Anthropic launched Claude Fable 5 and Claude Mythos 5 at $10 per million input tokens and $50 per million output tokens. Fable 5 leads FrontierCode among frontier models, while Mythos 5 reports about 10x acceleration in drug design and about 80% scientist preference in blinded molecular biology hypothesis tests.

Why it matters: HKR-H/K/R all pass: this is an official Anthropic dual-model release with pricing, coding benchmark, and drug-design speed claims. As a major Claude model update plus Anthropic substantive-update bump, it sits in the 85–94 band.

Take

Anthropic tied the capability launch to access control this time. Fable 5 goes to general users, while Mythos 5 starts inside Project Glasswing and trusted access. The hard detail is not the benchmark table. It is one base model with two gates: Fable 5 routes some cybersecurity queries down to Claude Opus 4.8, with triggers averaging under 5% of sessions.

The $10/M input and $50/M output pricing is less than half of Claude Mythos Preview, so Anthropic is preparing for real usage, not a museum-grade frontier demo. Stripe’s 50-million-line Ruby migration claim is wild: one day versus more than two months for a team by hand. I still treat that as customer PR until independent runs show the same pattern. Mythos 5’s security power arrives through a US government channel first; access policy, not API price, sets the adoption curve.

The Verge · AIJun 10, 01:00

Anthropic releases its first Mythos-class model Claude Fable 5

Anthropic launched Claude Fable 5, calling it the most capable model it has ever released widely. It excels at software engineering, knowledge work, and vision, with its lead growing on longer, more complex tasks. This is the first broad release from the Mythos family, previously deemed too dangerous because of cybersecurity capabilities. New safeguards that block responses in specific high-risk areas made the release possible. The post doesn't disclose benchmarks, pricing, or a launch date.

Why it matters: First public Mythos-class model from Anthropic, previously withheld for safety reasons. Long-horizon task gains and a new safety interceptor are concrete new info. HKR all hit; this is an industry-shaking release.

TechCrunch · AIJun 10, 01:00

Anthropic releases Claude Fable 5, a public version of its top Mythos model with hard safety limits

Anthropic opened its most powerful model family to the public for the first time with Claude Fable 5, a version of Mythos. It excels at software engineering, knowledge work, and vision, but blocks responses in high-risk areas like cybersecurity, biology, chemistry, and distillation, falling back to Claude Opus 4.8. Mythos was previewed in April for select partners only due to cybersecurity concerns. The post doesn't spell out Fable 5's parameter count, pricing, or regional availability.

Why it matters: Anthropic's first public release of a Mythos-tier model is a substantive product launch with explicit safety mechanisms. Cross-source cluster confirmed, all three HKR axes hit. Not scoring higher because the post doesn't disclose benchmark comparisons or pricing details.

Hacker News front pageJun 10, 00:58

System Card: Claude Fable 5 and Claude Mythos 5

Anthropic published a 319-page system card for Claude Fable 5 and Claude Mythos 5, stating that Fable 5 is for general use with biology and cybersecurity safeguards, while Mythos 5 lifts relevant safeguards and is limited to trusted partners starting with Project Glasswing.

Why it matters: HKR-H/K/R all pass: Anthropic documents two Claude 5 configurations, calls Mythos 5 its most capable model, and gives safety-gating details. This is a same-day Claude substantive update, placed in the 85–94 band.

Take

Anthropic turned this release into a two-lane product: Claude Fable 5 for general users, and Claude Mythos 5 with relevant bio and cyber safeguards lifted for trusted partners starting with Project Glasswing. That is a clean admission that frontier capability no longer ships safely through one uniform API surface.

The hard detail in the 319-page card is not “most capable model.” It is that Mythos 5 scores far ahead of Claude Opus 4.8 on cyber tasks, is treated at CB-1 but near the CB-2 line, and can significantly uplift well-resourced threat actors. METR’s read that AI R&D ability remains below Anthropic engineers keeps the runaway-agent story contained. The product move still says the quiet part loudly: access tiering is now part of model safety, not an enterprise packaging trick.

AI HOT (Curated Pool)Jun 10, 00:54

Apollo and Blackstone Team Up on $35 Billion AI Financing Deal

Apollo and Blackstone are working on a $35 billion AI financing deal involving Anthropic and Broadcom; the post says Wall Street is creating financing models for expensive AI chips, but it does not disclose the deal structure.

Why it matters: HKR-H/K/R all pass: the $35B size and Apollo/Blackstone plus Anthropic/Broadcom mix are strong. Deal structure, use of funds, and terms are not disclosed, so this stays in the 78–84 band.

Take

A $35B AI financing deal says less about Anthropic’s model lead than Wall Street turning GPU demand into structured credit. The visible facts are Apollo, Blackstone, Anthropic, Broadcom, and $35B. Bloomberg’s article body is blocked by a 403, so deal structure, collateral, tenor, and pricing are not disclosed.

I don’t buy the clean “AI boom gets funded” framing. Model labs have already pushed capital into training clusters, inference subsidies, and chip prepayments. CoreWeave showed the template: use GPU assets and cloud contracts to raise debt-like financing. Broadcom’s presence makes this look closer to ASIC or networking hardware orders being financed ahead of revenue. The risk is not whether capital shows up. The risk is whether Anthropic’s paid inference can carry long-duration capital costs without turning every token into a margin tax.

Jun 9Tuesday

Computing Life · Share · YageJun 9, 08:00

A rice blast experiment in Anthropic's Fable 5 safety report shows who AI can't replace

Anthropic ran a rice blast experiment in Fable 5's safety report: six biology PhDs paired with LLM experts used Claude Mythos 5 to design an agricultural pathogen defense in 16 hours. Two generalist teams beat all specialist teams—work the experts estimated would take 2–3 months manually. AI matched experts at literature search and cross-domain synthesis but repeatedly failed at judging whether an answer was correct or when to stop. The one variable never controlled for was the LLM expert—someone who knows the model fabricates citations, overestimates feasibility, and won't self-correct, and who stays in the loop to calibrate every output. Anthropic's conclusion that Fable 5 hasn't crossed the bioweapons risk threshold hinges on the assumption that ordinary users lack this person.

Why it matters: A controlled experiment inside Anthropic's Fable 5 safety report with counterintuitive results and high information density: generalist teams with LLM experts beat rice blast specialists. AI matched experts on literature search but repeatedly failed at judgment. HKR all hit, b...

Take

This piece is worth opening because it pulls a sharp angle out of a safety report. Anthropic paired six biology PhDs with LLM experts, gave them Claude Mythos 5, and asked them to design a rice blast defense in 16 hours. Two generalist teams beat all specialist teams—work the experts estimated would take 2–3 months manually.

AI won at literature search and cross-domain synthesis. It lost at judging whether an answer was correct or when to stop—fabricating citations, overestimating feasibility, and failing to self-correct even after detecting flaws.

But one variable was never controlled for: the LLM expert. This person wasn't writing prompts. They knew where the model would fail—hallucinated references, over-optimism, complexity bias—and stayed in the loop to catch every instance. Anthropic's conclusion that Fable 5 hasn't crossed the bioweapons risk threshold rests on the assumption that ordinary users lack this person.

I'd discount the sample size—3 vs 3, and the designs were never lab-tested. Don't read this as "generalists replace experts." The useful bit is the framework: ask how much of the work is finding answers vs judging them, and whether you're the one who can fill the gap.

Computing Life · Share · YageJun 9, 08:00

Fable 5 Is Expensive, but Anthropic Published the Cost-Saving Answer Two Months Ago

Anthropic launched Claude Fable 5 at $50 per million output tokens, over 3× the price of Sonnet 4.6. The cost-saving answer shipped in April: the advisor tool lets a cheap model do the work and calls Opus for a few hundred tokens of advice only when stuck. Sonnet + advisor scored 2.7 points higher on SWE-bench than Sonnet alone while costing 11.9% less. Fable 5 currently only advises itself, but its pricing makes the advisor role the only comfortable fit. The post recommends testing Fable 5 in Claude Code before the June 22 free window ends, then running three API configs: Sonnet solo, Sonnet + advisor, Opus solo.

Why it matters: Fable 5 launch is big news, but this piece is a second-order take on the advisor tool as a cost-saving pattern, not a first-party release. HKR all hit: pricing contrast creates curiosity, advisor mechanics are concrete, and the cost decision hits agent builders directly. But i...

Take

This piece clicks because Fable 5's pricing turns "should I use the strongest model" into a real budget question. At $50 per million output tokens — over 3× Sonnet 4.6 — running it end-to-end in an agent loop is hard to justify.

The fix shipped quietly in April. The advisor tool lets a cheap model like Sonnet or Haiku run the whole task, calling Opus for a few hundred tokens of advice only when it's stuck. The cheap model keeps the steering wheel. Anthropic's own numbers: Sonnet + advisor scored 2.7 points higher on SWE-bench than Sonnet alone, while costing 11.9% less. Better plans meant fewer retry rounds, and the advisor cost got paid back by saved executor turns.

Right now Fable 5 can only advise itself — the combo everyone actually wants, Sonnet executing with Fable 5 advising, isn't open yet. But cross-tier pairing has precedent: Opus 4.8 quickly appeared as an advisor option for cheaper models after its May launch. Fable 5's pricing is practically built for the advisor slot: high unit price times tiny token volume is the only comfortable fit. Until that combo opens, you've got a free window in Claude Code through June 22 to test Fable 5's ceiling.

Evaluate via direct API, not Claude Code integration — a token double-counting bug has been open for five weeks.

Computing Life · Share · YageJun 9, 08:00

Paying Fable's Price, Getting Opus's Goods: The Other Identity of AI Safety Guardrails

Anthropic launched Fable 5 and Mythos 5 on June 9—same base model, two deployment tiers. Fable 5 is public but a classifier hands off high-risk queries to the older Opus 4.8. Mythos 5 is uncapped and restricted to trusted partners. The article connects this to four other moves in 30 days: splitting programmatic usage from subscriptions, cutting fast-mode premiums, filing an S-1, and pulling Fable 5 from subscriptions on June 23. The argument: Anthropic is using safety mechanisms as price fences, letting high-value users self-select and pay more. The classifier re-sorts every request in milliseconds. The safety function is real; the pricing function is also real. A three-part test for spotting a price fence is offered: a quality or access gap, a price gap, and a self-selection mechanism.

Why it matters: Reframes Fable 5's safety classifier as a pricing fence, threading five product moves across 30 days into a coherent commercialization narrative. Concrete timeline and pricing data back it up. Docked slightly because density drops in the second half and the thesis is debatable...

Take

This piece clicked for me because it connects five Anthropic moves in 30 days into one pricing logic: splitting programmatic usage from subscriptions on June 15, cutting fast-mode premiums from 6x to 2x on May 28, filing an S-1 on June 1, the Fable 5 / Mythos 5 dual launch on June 9, and pulling Fable 5 from subscriptions on June 23. Individually they're routine product updates. Together they're the same thing: slicing model usage into separately billable dimensions.

Fable 5 and Mythos 5 share the same base model. The difference is entirely in deployment. Fable 5 runs every request through a classifier—if it detects cybersecurity, biochemical, or distillation intent, the query gets handed off to the older Opus 4.8. Mythos 5 removes those guardrails but is restricted to trusted partners. The article frames this with the economics concept of a price fence: like railway companies leaving third-class carriages roofless not to save money, but to make second-class passengers self-select out of the cheap option.

I buy this framework because three hard facts back it. One, the safety function is real—the system card rates the model's CB-1 bio capabilities and admits the assessment is "far less clear than previous models," and UK AISI already made progress on a universal jailbreak during preliminary testing. Two, the classifier re-sorts every request in milliseconds, moving the fence from the pricing page into the inference loop. Three, the organizations that need uncapped cybersecurity capabilities are exactly the ones that can sign enterprise contracts, pass vetting, and accept data retention terms—trust and willingness to pay track each other in this market.

One thing the article doesn't unpack: Anthropic says the classifier triggers on less than 5% of sessions, but doesn't give a scenario breakdown. If that 5% clusters in a few high-frequency security use cases, the fence is even cleaner for regular users. Also, once Fable 5 leaves subscriptions on June 23 and users pay API rates of $10/$50 per million tokens for usage credits, the real question is how much heavy users' bills jump—the post doesn't run those numbers, and that's the figure to watch next.

Financial Times · TechnologyJun 9, 07:50

Apollo and Blackstone Raise $35bn in Chip Financing Deal for Anthropic

Apollo and Blackstone raised $35bn in a chip financing deal for Anthropic, and the RSS snippet says the transaction supports the Claude maker’s AI growth plans.

Why it matters: HKR-H/K/R all pass: FT reports a $35bn Anthropic chip-financing deal involving Apollo and Blackstone. The article lacks term, cost, and procurement detail, so it sits in the lower 85-94 band rather than higher.

Take

Anthropic’s $35B chip financing says the Claude fight has moved from model quality into balance-sheet engineering. The RSS snippet names Apollo and Blackstone and says the deal funds Anthropic’s AI growth plans; it gives the headline number, but not rate, tenor, collateral, GPU ownership, or lease-versus-debt structure.

This smells less like a clean funding round and more like private credit turning AI compute into a packaged asset class. OpenAI leaned on Microsoft and Oracle, and xAI made Colossus a campus-scale buildout; Anthropic is now using financial machinery to chase the same compute curve. My concern is simple: $35B buys throughput, but it also pins down gross margin. If Claude cannot convert enterprise API demand and agent workloads into durable usage, the financing terms will bite before the model architecture does.

AI HOT (Curated Pool)Jun 9, 06:50

Altman Says OpenAI Has Entered Its Third Phase: Making AI Widespread, Easy to Use, and Safe

OpenAI said on Monday it has entered its third phase, naming three goals: automated AI researchers, faster economic growth, and personal AGI for everyone, while calling for an international body to manage AI risks.

Why it matters: HKR-H/K/R all pass: OpenAI’s “third stage” and personal AGI frame give it a hook, with three goals and an international-agency proposal. No model release, timeline, or measured capability is disclosed, so it stays below 85.

Take

OpenAI’s “third phase” is not a softer mission statement; it is an IPO-era bundle of scale, automated AI researchers, personal AGI, and global oversight. The article gives three targets: automated AI researchers, faster economic growth, and a personal AGI for everyone. It also says OpenAI quietly filed for an IPO the same day, while admitting the listing is still far off.

I don’t buy the clean “accessible, easy, safe” wrapper. OpenAI says frontier development may need pauses, then names abundant supply and lower cost as core stage-three work. That is a permission structure for more compute, cheaper inference, and wider distribution. Anthropic’s recent pause language reads more like a boundary-setting move; OpenAI’s version reads like a capital-market filing with safety attached.

AI HOT (Curated Pool)Jun 9, 06:24

Claude Supports Apple Foundation Models Framework With New Swift Package

Anthropic released a Swift package that lets Apple developers call Claude inside the Foundation Models framework with three lines of code, returning typed Swift values and handing off multi-step reasoning, code generation, web search, and data analysis on iOS 27, macOS 27, and related platforms.

Why it matters: HKR-H/K/R all pass: Anthropic is shipping a concrete Claude Swift package for Apple Foundation Models, but this is a developer integration rather than a model release, so it sits high in the 78–84 featured band.

Take

Anthropic is making the right platform bet here: Claude is not chasing another chat surface, it is entering Apple’s developer framework. The concrete hook is unusually clean: a Swift package, three lines of code, typed Swift values, and support for multi-step reasoning, code generation, web search, and data analysis across iOS 27 and macOS 27. For an Apple developer, that feels closer to native plumbing than a separate REST client.

I don’t buy the implied jump from “easy to call” to “developers will default to Claude.” The article gives no pricing, latency, model-selection rules, or cloud-versus-on-device handoff behavior. It also does not say where Claude ranks against Apple’s own Foundation Models inside the framework. Anthropic gains distribution surface here, not default status.

AI HOT (Curated Pool)Jun 9, 05:51

OpenAI confidentially files for IPO as Anthropic enters capital race

OpenAI filed a confidential S-1 with the SEC to start IPO review without public revenue or loss data; Anthropic filed last week, and Sam Altman said AI will handle a large share of OpenAI research by March 2028.

Why it matters: HKR-H/K/R all pass: dual frontier-lab IPO filings and Altman’s March 2028 research claim are major. Thin sourcing from an X post keeps it at 90, below the 95+ IPO band.

Take

OpenAI’s confidential S-1 moves the frontier-lab fight onto the balance sheet. The filing lets it start SEC review without exposing revenue, losses, or customer mix; Anthropic filed the same way last week. That timing smells less like coincidence and more like both labs needing public markets to finance training clusters, inference subsidies, and talent.

Altman’s claim that AI will handle a large share of OpenAI research by March 2028 should not be read as a clean technical roadmap. IPO investors will ask when the burn curve bends. “Automated AI researcher,” scientific productivity, and personal AGI all package R&D labor as a compressible cost line. The catch: the snippet gives no revenue or loss data, so the market is being asked to price the story before seeing operating leverage.

AI HOT (Curated Pool)Jun 9, 05:38

OpenAI Files for IPO After Anthropic

OpenAI confidentially filed an S-1 with the SEC on Monday to start its IPO process; Anthropic filed the same type of document on June 1, and the post does not disclose executive compensation or business-risk details.

Why it matters: HKR-H/K/R all pass: OpenAI confidentially filed an S-1, and Anthropic filed on June 1, matching the top-band example of a foundation-model company IPO. Missing pay and risk details trim little from the event weight.

TechCrunch · AIJun 9, 05:29

OpenAI files confidentially for IPO, following Anthropic

OpenAI filed confidentially for an IPO a little more than one week after Anthropic filed to go public; the post does not disclose valuation, fundraising size, or listing timeline.

Why it matters: HKR-H/K/R all pass: OpenAI’s confidential IPO filing is industry-shaking and follows Anthropic by just over a week. Valuation, raise size, and timing are not disclosed, so it stays at 95.

Take

OpenAI filing now is less about IPO readiness than forcing Anthropic’s timing into a capital race. The article gives one hard fact: OpenAI filed confidentially a little over one week after Anthropic. It gives no valuation, raise size, or listing timeline. Any chatter about a trillion-dollar valuation or a specific float size belongs in the rumor bucket until the S-1 appears.

I read this as compute capex pushing both labs toward public-market money. OpenAI’s training and inference bills are no longer a clean fit for private rounds, and Anthropic is in the same bind. The difference is OpenAI carries nonprofit governance baggage and a deep Microsoft commercial tie. Once the filing is public, investors will not price AGI poetry; they will price revenue quality, gross margin, and contracted compute obligations.