The Chopping Block: Bitget's 387 Million Dollar Hack, Kalshi's Cooked Perps Volume, and Agentic Bank Runs

Bitget lost about $387 million in a hack that speakers said likely involved hot or warm wallets, though the infrastructure details were unclear. They considered North Korea’s Lazarus Group the very likely culprit. Bitget reportedly had a $460 million insurance or user-protection fund, withdrawals had begun, and speakers described the exchange as profitable and solvent, with no contagion from the incident. The attackers reportedly moved stablecoins toward Ethereum and then used THORChain to transfer assets across chains, a pattern speakers said often follows major hacks.

The incident sharpened a debate over how much discretion crypto protocols should use to block suspected illicit funds. NEAR Intents’ Shield reportedly froze about $500,000 and refused to route more than $100 million in attempted trades, screening for anomalies and hack-related signals at the validator layer; THORChain permits cross-chain spot trading without comparable restrictions. One speaker favored NEAR’s mix of decentralization and safeguards, while another argued protocols should be either consistently neutral or deliberately censoring, rather than intervening case by case. The discussion framed the choice as a trade-off among protection, privacy, liquidity, and convenience: Coinbase represented high control and Uniswap no discretion, with NEAR and THORChain between them. Speakers also cited Arbitrum’s earlier decision to route stolen funds into a governance process to seize them, and contrasted Ethereum’s inability to shut down with THORChain’s roughly month-long shutdown after a hack, which they attributed partly to its small validator set.

A separate Consensus incident exposed a compromised Ethereum validator but did not reportedly cost users their staking principal. The operator exited the stake and rotated keys; less than 1 ETH in unclaimed validator rewards was exposed, while affected stake could forgo about 200 ETH in rewards. Speakers said about 5% of Lido stake was delegated to Consensus validators and debated whether lost yield should add a meaningful risk premium. In one hypothetical, a provider holding 5% of stake with a 5% annual chance of a one- to two-week rewards interruption would imply roughly a one-basis-point impact across the stake. Speakers cautioned that this mild interruption scenario did not cover malicious slashing or principal loss, and disagreed over whether Ethereum would roll back after catastrophic mass slashing.

Benny alleged that Kalshi’s perpetuals volume was inflated by market-maker incentives that encouraged wash-like trading. He pointed to repeated $5,500 trades in public order books and roughly $3 million in reported open interest against hundreds of millions of dollars in daily volume, including one speaker’s $500 million figure. The gap led speakers to suspect activity was dominated by market makers rather than organic demand, but they could not verify the allegation or Kalshi’s defense that the trades were genuine without underlying data. The Wall Street Journal reported that the CFTC was investigating Kalshi and some of its volume claims, and Kalshi said it was ending its liquidity-incentive program. Speakers stressed that rebates are common on traditional exchanges and are not inherently improper; one speaker’s estimate that as much as 98% of Kalshi volume might be fake was not established by the discussion.

An Apollo slide prompted discussion of whether AI agents could create “agentic bank runs” by shifting idle checking or sweep-account balances into higher-yielding products such as Treasury bills. Speakers distinguished automated rebalancing from a run: banks could respond to deposit outflows by raising rates, but that would reduce returns. Sticky consumer deposits currently let banks pay below prevailing market rates and support net interest margins. One view was that agent-driven transfers would grow gradually with adoption and wealth transfers; another warned that agents making similar decisions could accelerate an SVB-style run. In a widespread-adoption scenario, one speaker said a roughly $10 billion “hot ball of money” could grow to $3 trillion.

Consumer reluctance to share financial information and resistance from established banks could slow that shift. One speaker cited JPMorgan’s roughly decade-long fight over Plaid’s read-only account access, while Column, Mercury, and AeroBore were cited as newer banks offering APIs or MCP hooks with fine-grained permissions. Agents that identify overcharges or cancel difficult subscriptions could also weaken consumers’ reliance on inertia. The debate left open whether such tools would mainly improve individual financial choices or, when many users act alike, make deposit flows faster and less stable.