Anthropic's Crypto Research Went to the Top of Hacker News by Publishing the Exact Dollar Cost and Its Own Model's Failure Mode

The Play

Anthropic's cryptography research post didn't lead with a breakthrough claim. It led with a receipt.

The post describes two results: an attack on HAWK, a NIST post-quantum signature candidate, and an improved attack on round-reduced AES. Both are genuinely notable — HAWK had "survived two rounds of expert human review over a period of two years" before Claude Mythos Preview found a way to cut its "key strength in half," and the AES result improved on "the previous best attacks by 200-800×."

But the growth mechanic isn't the results themselves. It's what Anthropic chose to disclose alongside them.

What Anthropic disclosed that most companies wouldn't

Most capability research posts report the win and leave the inputs vague. Anthropic did the opposite:

• The exact cost: "Each of the two results cost roughly $100,000 in API cost to develop."
• The exact time: HAWK took Mythos "just 60 hours of work."
• The scope limit, stated plainly: "neither of these results has a practical impact on today's computer systems; no production software will have to change as a result."
• The near-failure: a direct quote that "the models tend to think it is impossible to solve so they don't try" — an admission that the default behavior almost prevented the result entirely.

That combination — cost, time, limits, and a specific failure mode — is a different kind of proof than a benchmark score. It's an itemized receipt a reader can check against their own assumptions about what AI-assisted research actually costs and where it actually struggles.

Why it worked

The post reached 143 points and 74 comments on Hacker News. That's not a viral explosion — it's the size of thread you get when the content is specific enough to argue about but narrow enough that only people who actually read it weigh in.

Compare that to a typical capability announcement: a headline claim, a demo, and no real numbers behind it. Those posts get a quick upvote and no real discussion, because there's nothing concrete enough to check. Anthropic's post gave readers an actual dollar figure, an actual hour count, and an actual admitted limitation to test against their own understanding — and testing claims is what drives comments.

The growth play to steal

You don't need a frontier research lab to use this. The transferable move is: when you publish a result, publish the unglamorous inputs next to it.

• state the real cost to reproduce your result, not a vague investment claim
• state the real time it took
• state where your own approach almost failed or fell short, in a specific, quotable way
• name any independent collaborators or validators attached to the work

A result with its receipts attached reads as something a stranger can verify. A result without them reads as a claim they're asked to trust — and claims without receipts get scrolled past, not argued about.

Bottom line

Anthropic's growth mechanic here wasn't the cryptography. It was publishing the $100,000 cost, the 60-hour clock, and the model's own near-failure right next to the win. That specificity is what turned a capability announcement into a 74-comment technical discussion — and it's a move any team publishing research or case-study results can copy directly.

Source: https://www.anthropic.com/research/discovering-cryptographic-weaknesses