The room now has a much cleaner separation between exposure, installation, execution, and compromise. For the 575-plus reported OpenClaw skills, catalog presence or download does not establish that code ran. Installation evidence supports disabling and removing the skill; invocation logs, module loads, child processes, filesystem changes, secret access, or attributable traffic are needed to establish execution. Isolation, rebuilding, and credential rotation should therefore follow demonstrated execution, integrity changes, secret access, outbound communication, or telemetry gaps that prevent those outcomes from being excluded. TeamPCP raises a different concern because executable Python .pth content may run through normal interpreter startup and survive ordinary package removal. The material presented here, however, does not complete the requested SleepyDuck assessment, so that portion remains open rather than inferred.
On Ankr, approximately $304,500 across five withdrawals is supported, with contemporaneous estimates varying from about $292,000 to $304,500. The important finding is that materially altered withdrawals passed trusted-signer authorization while referring back to one previously redeemed, very small Optimism deposit. That supports an authorization-binding and one-time-redemption failure. It does not yet identify the root cause as signer-key theft, collusion, weak domain separation, or a particular verification defect. Those remain competing explanations, and this was not a simple byte-for-byte replay.
Operationally, SSHole stays first in the queue, but with narrower claims than the headline suggests. The first four hours are about preserving SSH and cloud evidence, inventorying exposed OpenSSH systems, prioritizing confirmed musl and PAM-disabled configurations with the relevant password marker, and restricting access through approved jump hosts. Reported exploitation and root access on an Alpine honeypot do not prove reliable bypass across every musl/OpenSSH configuration or password-disabled root account. Cloud trust paths, high-risk iOS users, and refrigeration systems also require evidence-led containment rather than broad shutdowns or attribution from exposure alone.
The next question is where actor attribution outruns that same evidence discipline. CVE-2026-0257 exploitation is high-confidence but actor-neutral; one reported chain provides moderate support for a Qilin affiliate, while the “15%” figure reflects market or leak-site prevalence, not an initial-access method. We now need to pressure-test whether the available record supports anything stronger than an isolated affiliate-level linkage.