ARCHIVE · DAILY SELECTIONS
Selections for August 28, 2026
BitShovel tracked 28 signals that day; start with the 3 focus picks. This page is an immutable record of the choices made then — useful for checking them against what happened next.
Trail of Bits tested it: a cyber-capable agent escaped its VM three ways
A Trail of Bits engineer with preview access to GPT 5.6-Cyber (via Patch the Planet) gave it a real challenge: escape the QEMU/KVM VM (Debian 12, AMD Zen3) he uses for sandboxing. It succeeded three times — first via newly disclosed host-kernel bugs, then via disclosed bugs not yet classified as security issues after full updates, and finally by finding several 0-days in freshly rebuilt upstream sources — running autonomously for hours, pulling papers and writing oracles and minimal examples on its own.
Why it was selected"Just put it in a VM" is the de facto default for running agents today, and this first-hand report falsifies it — required reading if you sandbox agents or evaluate vendors' isolation claims.
Caveat[Evidence provenance: Trail of Bits' own blog] One machine, one demonstration — not a systematic benchmark; it targets a cyber-specialized model, and Lobsters upvotes are attention, not proof that the method generalizes to ordinary models or every configuration.
Anthropic opens real Claude conversations to self-directed independent research
Anthropic's Aug 26 post Enabling Independent Research describes a spring pilot in which three external institutions designed their own studies through Anthropic Insights (formerly Clio), its privacy-preserving analysis tool, and independently analyzed aggregated real-world Claude conversations; results are public. The company frames it as the first such public independent studies on an AI company's own usage data and invites more researchers via an interest form.
Why it was selectedReal-world usage data has been concentrated inside a handful of labs; this is the first crack opened by the incumbent itself — a structural shift worth tracking for anyone interested in AI's societal impact or independent verification.
Caveat[Evidence provenance: Anthropic's own announcement] Not an independent review: the description is company self-reporting, the pilot covers three institutions with the platform running collection — attention around it is attention, not proof of genuine independence.
New on Product Hunt: ChatCut Desktop
New Product Hunt feed entry (50 entries in the current feed). Community-upvoted into the daily selection; captured by BitShovel through the official feed.
Why it was selectedThe Product Hunt feed is the first stop for international new-product attention; fresh listings belong on the radar to watch retention.
CaveatThe listing and copy are a creator self-report, not an independent review; launch attention is not proof of retention, and Product Hunt is spotty from mainland China.
28 tracked signals that day