@Ondo says anthropic is coming onchain on ipo day.
the biggest ai listing of the year, tokenized from session one.
read the wrapper before the trailer.
it is not a share.
ALT Card comparing three Anthropic products: a live HIP-3 perpetual, an Ondo structured note planned for IPO day, and the actual Nasdaq share. Only the share sits on a register.
two other anthropic-shaped things are easy to confuse with this.
a hip-3 perp on anthropic already trades (@entropyIO's ANTH, and @tradexyz has ANT). that's leveraged synthetic exposure, not a note and not a share.
when anthropic lists, $ANT on nasdaq is the actual stock.
so on day one you could theoretically touch three products with the same name on them:
◦ a perp
◦ a structured note
◦ the share itself
only one of those puts you on a register.
my read: ipo-day tokenization is a distribution story. the wrapper is the product.
ondo did not say anthropic is a partner, and the launch depends on the ipo actually happening.
if you care about voting, information rights, or u.s. access, this is not that rail.
if you care about getting economic exposure into a wallet the day a name starts trading, this is exactly that rail.
pick the rights before you pitch the ticker.
agents didn't break rsa.
@GoogleDeepMind published two working papers on friday: human-guided agents stress-tested leftover 1024-bit rsa.
the papers say modern 2048-bit and larger deployments are not threatened.
the story is speed, not a break.
ALT Card explaining DeepMind RSA-1024 research numbers: 1600x is an estimate never run at size, 3.5x was measured, 30x is per extra key, and padding stops helping at half the block. 2048-bit RSA is not threatened.
what this does not say.
▸ no 1024-bit key was factored. factoring that size is still estimated at 500k–1m core-years
▸ 1024-bit is already not approved for generating new signatures
▸ every 2048-bit cost in the papers is asymptotic
▸ the model needs direct signing calls on caller-chosen inputs, made before the target was known
▸ leftover 1024-bit keys and raw signing interfaces are the thing to inventory. not your bank's 2048-bit tls
two ai agents wrote 33,500 lines, ran 336 experiments, and burned 391 core-years on a published rsa-1024 problem.
the human sent about 20 messages. most of them were approvals.
the paper's own conclusion: modern 2048-bit rsa is not threatened.
a small number from @AnthropicAI's memory post that coding-agent people should actually use.
claude code's auto memory only injects the first 200 lines, or 25kb, of MEMORY.md into a new session.
the rest is an index. claude has to go look it up.
they call the offline cleanup "dreaming": prune the top layer, leave links behind, correct notes that drifted.
longer instruction files reduce adherence. the top of the pile has to stay small on purpose.
the skill i want from coding agents too.
@OpenAI's new biobasicsbench (w/- ligo): 100 tasks from real wet-lab experiments.
not "did the model get the answer." did it read the raw data, catch errors, and notice when the evidence doesn't support the conclusion.
that's the same failure mode i see when claude and codex fight over a codebase: fluent, confident, and sometimes unwilling to say "we don't know yet."
A new way to measure how AI agents handle the routine analyses of wet-lab science—BioBasicsBench.
Developed with Ligo, a biotech company focused on enzyme design, the benchmark draws on 100 tasks from real experiments to test whether agents can interpret raw lab data, identify errors, and recognize when the evidence doesn’t support a conclusion.
Accelerating biological research starts with getting the everyday work right. BioBasicsBench shows where models need to improve, and helps us track what those gains mean for scientific progress, biosafety, and biosecurity.
openai.com/index/measuring-h…
the better memory folder was the smaller one.
@AnthropicAI's new note on agent memory: claude "dreams" between sessions — it rereads logs, corrects notes, and rebuilds what the next session will need.
in a pokémon eval, at about 10,000 steps:
▸ sonnet 3.7 wrote 413 files across 225 folders, earned 1 badge
▸ fable 5 wrote 9 files + a global.md, earned 4
sonnet treated everything as worth saving. fable kept plans and tested rules, not what the screen already showed.
memory is a prediction about what the future will need. more files isn't more memory.
Memory is what allows Claude to learn from experience. In a new Engineering Blog post, we share how “dreaming” helps Claude organize what it’s learned, correct its mistakes, and find what it needs more easily the next time. anthropic.com/research/compo…
not a share. a structured note, planned for ipo day.
@Ondo says anthropic is coming onchain through ondo stocks when it lists.
their own docs are clear:
▸ issued as a debt instrument by ondo global markets (bvi)
▸ economic exposure, redeemable for value
▸ no shareholder voting or information rights
▸ not offered to u.s. persons
no listing date yet.
the scarce part isn't "anthropic onchain." it's which rights the wrapper actually carries.
hip-3 is ten dexes on paper. on a sunday morning it's basically one quiet book.
pulled the hyperliquid hip-3 registry this morning (7:51 am ist).
▸ @tradexyz: $368m / 24h (91.8% of hip-3 volume) · $4.28b oi (95.9%) · 120 live markets
▸ @entropyIO: $31.9m / 24h (8.0%) · $166m oi
▸ paragon: $0.9m
▸ markets by kinetiq: $0.1m
▸ the other six dexes: zero live markets
whole hip-3 book: $401m volume across 168 live markets (325 names in the registry).
this 24h window starts after friday's u.s. cash session. yesterday morning's pull still had friday in it — skhx alone printed $535m. today the entire xyz book did $368m.
oi barely moved: $4.28b on xyz vs $4.29b yesterday. the positions stayed open. the tape went to sleep.
hip-3 still runs on the tradfi clock.
ALT Chart of HIP-3 24-hour volume share on November 8, 2026 at 7:51 AM IST: trade.xyz 91.8 percent, entropy.io 8.0 percent, with total HIP-3 volume of $401 million after the U.S. Friday cash session.
this week made the tokenized-stock split impossible to ignore.
three rails shipped headlines at once.
▸ dtc entitlement twin
▸ issuer-sponsored share
▸ rfq economic claim
same label. different rights.
ALT Three rails of tokenized stocks: DTC twin, issuer register, RFQ claim
rail 3 — rfq economic claim.
@jtx_trade equities went live nov 5 w/- @Ondo + @Backpack on solana.
▸ 1,650+ tokenized equities
▸ direct issuer quotes (rfq), not amm pools
▸ self-custodial + equity research tools
▸ generally not available to u.s. clients
ondo's own docs: economic exposure, not shareholder rights.