The Daily Token

Brute Force and Book Keys: Claude Fable Decodes a 1653 Royalist Cryptogram

Vals AI tasked Anthropic's Claude Fable 5.1 with unprompted cryptanalysis, leading it to solve Sir Thomas Urquhart's 370-year-old Cyphral Distich by indexing directly into the text's own structural paragraphs rather than applying classical frequency analysis. The result is a humbling reminder of human oversight—the solution required no sophisticated math, only methodical patience, though reliance on automated search runtimes risks substituting genuine conceptual breakthroughs for brute-force persistence.

Astra and Fable Stick to 2025 Benchmarks as Alignment Evals Stagnate

By continuing to test frontier models on minor iterations of 2025 alignment frameworks, major labs risk optimizing for legacy safety metrics while subtle, out-of-distribution failure modes go unmeasured. The reliance on familiar tests highlights a growing tension between rapid model scaling and the tedious work of inventing rigorous, modern evaluation paradigms.

Data collected by cars and sold to third parties

AI Robots – When will they be in our homes

Show HN: Dllog – Replay debug logs when an operation fails

Show HN: I made an automated day-by-day itinerary organizer for my trip to Spain

Device Drivers lab exercise – COSC562

Garry Tan wants US open-weight AI labs to 'distill' frontier models, too

Airbnb Preventing the Use of BnB

When Automated Scanners Blindly Follow CNAMEs into the Wild

An automated enterprise security scanner managed by Assetnote mistook a volunteer pool NTP server for internal Tesla infrastructure due to a raw CNAME delegation. The resulting 50,000 exploit probes sent to a hobbyist's IP illustrate the fragile boundary between automated exposure management and unauthorized network attacks.

Social Engineering Bypasses Revolut's Compliance Rails via Fake Legal Demands

Attackers forged government requests to extract customer data, exposing a vulnerability in manual compliance processes where trust is delegated to unverified paperwork. The breach highlights how operational automation often collapses at the exact boundary where software meets human bureaucratic discretion.

Anthropic Claims Houthis Employed Claude Code for Missile Guidance Development

The incident highlights how command-line agentic tools lower the barrier for technical execution in sensitive domains, forcing labs to confront the limits of automated safety guardrails once code leaves their hosted API environments.

Scraping Operation Exposes 7.3 Million Chess.com User Records

A massive data pull highlights how exposed public endpoints continue to be harvested at scale, reminding engineering teams that basic access controls matter far more than perimeter defenses. While no financial credentials were taken, the incident underscores the growing risk of structured scraping turned breach.

Lawyer cites fake witnesses in murder case and blames ChatGPT