Brian Tolzman

Agents are central to the future of the agentic economy. But the question is not whether an agent can make a mistake. It is what happens when a mistaken agent can sign. A trading agent reads token metadata, tool output, pages, and other agents’ messages, then acts. Those strings were not part of your prompt. Models still do not reliably tell data from commands. Giving the agent permission to trade does not make that text trustworthy. If the payload lands in CLAUDE.md or a hook, the agent will read it again tomorrow. @agentwormhole is trying to sit in front of the signature. read → check → verify → sign → pay. ➥ The actual problem ⛩ Prompt injection against an agent with a wallet is a transfer. A payment to an attacker’s address simulates perfectly. Correct balances. No revert. The model sees nothing wrong. Wallets answer what the transaction will do. Almost none answer whether it is the transaction that was asked for. That is why this exists. Not an AI security meme. A check outside the model, before the money moves. ➥ Three surfaces ⛩ → Instructions. wormhole-guard is free, offline, on PyPI. It watches writes to CLAUDE.md, .cursor/rules, hooks. The June 2026 pattern that matters here is persistence: a payload that writes itself into the file the agent re-injects next session. Hash the baseline. Refuse the write. You do not need to hold to run this. → Payments. wormhole-x402 compares the unsigned payment to the merchant quote on a channel the model never touched. Wrong destination, wrong amount, sneaky delegation → refused. Offline. No extra RPC. Solana and EVM. → Token launches. Name, ticker, description are cheap to poison. Zero-width characters, hidden instructions, impersonation. The launch layer scans at creation on Robinhood Chain, on demand on Solana, attests the exact bytes, and badges the token. The badge says checked, changed, or findings. It does not say safe. ➥ What they are not claiming ⛩ They published their own miss. Guards that trust a tool’s name get walked around if the other side renames the tool. They measured it, patched it, left the paper up. They also say they do not stop slopsquatting at npm install. They catch persistence after the payload tries to stay. A hole in Azure’s control plane is a different layer. They said they would have done nothing there. Layers, not a silver bullet. That is the part of their timeline worth keeping. ➥ The loop ⛩ Paid checks are meant to mint a micropayment. → Fees buy and burn it. They have reported 642,645 scans, 37,413,789 burned, 3.74% of supply removed and about $9,639.67 spent on buybacks. (keep in mind that these numbers might have changed) Scans are not paid calls. Many tools are free. Per-check fees are cents. Meaningful burn needs volume. They are paired with $PLTR: security token next to a security-related equity name, on Robinhood Chain. Easy narrative. Not the same as being the default check inside Robinhood’s trading MCP. ➥ What I am watching ⛩ Whether agents call the check before sign, not after. Whether x402 volume actually needs quote verification. Whether launchpads attest metadata before an agent reads it. Whether buybacks track usage and become a strong flywheel. Whether the next agent-loss incident makes a pre-sign check mandatory. If models, wallets, and MCP hosts ship this natively, a bolted-on layer gets thinner. If they do not, someone has to own the channel between untrusted text and a signature. Your agent can read the internet. The internet should not get to spend your money. You can read more about them at: agentwormhole.com Like always, NFA and DYOR fam. FYI, was shared in the Dojo private Telegram group at around a $400K market cap, then around $1M to $2M in the public Telegram channel.

India's biggest IPO lists tomorrow. Trade it today. Confidentially.
 $JIO pre-IPO perp is live. Nobody can trace your position back to you.
 Only on perps.near.com. Powered by @HyperliquidX & @tradexyz

Great summary of ethereum:0xca14007eff0db1f8135f4c25b34de49ab0d42766 from @Karamata2_2 I took a small position on Saturday at this price. I’m waiting for a pullback to $0.07 to build out my full position, or even $0.05 if it drops that low. Volume profile looks really good at these levels @Starknet is positioning itself nicely for the next run imo

dips are opportunities to buy lower, getting better entries. enjoy them

🔥 $STRK is trading around $0.10x - up more than 2x since my call, with a market cap of around $800 million. 1/ But I don’t think the interesting part is the pump. It is what shipped underneath it: - Starknet v0.14.4 is now live. - SNIP-36 can prove up to 1.1B L2 gas of computation in a single transaction, basically allowing much heavier logic to be compressed into one proof. => That matters for the privacy thesis because Starknet is no longer just talking about it. - The stack is already becoming usable: + strkBTC can be shielded, transferred privately and staked through Endur + Private DeFi is expanding into swaps, lending and yield + Privacy tooling is already open for builders StarkNet 2/ And now there is another optionality: Starknet becoming an L1 built around post-quantum security. That idea is still under discussion, not a roadmap commitment, but it gives the market another reason to rethink what Starknet could eventually become. 3/ But we wouldn’t ignore the risks. $STRK has moved much faster than the underlying usage. It gained more than 300% over four weeks while DeFi TVL increased only around 5% over a similar period. And roughly 127M STRK is scheduled to unlock around Oct 15, about 1.7% of circulating supply. So this is no longer the easy ~$0.05 call. The next test is whether strkBTC, privacy apps and DeFi activity can grow fast enough to absorb both the higher valuation and new supply. If they do, I think this re-rating has more room. I’m still holding my $STRK bag. Some chads are bullish at $STRK: @MoneyPrinter0x @CookerFlips @WisdomMatic @ryandcrypto @BitGod21 @notEezzy @ika_xbt @jedifractal @The__Solstice @astekz

The near:native chart looks very interesting at these levels. Let’s see if we can break resistance this time, at around $5.50 and start holding above it! 🔥 My spot target remains unchanged: I’m still aiming for double digits price and around the $15–$20 range.

the top 25 pumpfun deployers: > launched 770k tokens > extracted $128M from the trenches > produced only 6 tokens with a 20M+ MC that's $128M extracted just by 25 devs