CP SignalTechnology worth your attention.

Saturday, 17 October 2026

The Daily Edition: 5 stories, ranked by significance

  1. Artificial intelligence

    Claude reveals ECDSA curve origins

    Anthropic reports that an internal Claude research model solved a long-standing open question in cryptography: the origin of the mysterious seed values NIST used to generate ECDSA elliptic curve parameters in 2000. Using pattern recognition and lattice reduction (LLL), Claude determined the seeds came from a 1990s C library PRNG (lrand48) initialized with Unix timestamps from 1997-98, rather than a hashed passphrase as previously rumored, and also traced some curve base points to a separate RNG used in Arjen Lenstra's FreeLIP library.

    Read the storyNo. 1 of 5Anthropic NewsFri 16 Oct
  2. Artificial intelligence

    Designing more efficient quantum circuits with Claude

    Anthropic reports that Claude, given plain-language specs for eight quantum circuit design tasks and an automated correctness/efficiency evaluator, worked in a closed loop without human intervention to discover circuit designs using fewer qubits and gates than the best previously published algorithms. Highlighted results include a new modular inversion approach for elliptic-curve arithmetic and a roughly 12x cheaper version of the costliest step in a leading quantum optimization algorithm.

    No. 2 of 5Anthropic NewsFri 16 Oct
  3. Artificial intelligence

    Approve What, Exactly? Effect Binding and Trusted Rendering for Agent Step-Up Authorization

    A research paper defines twelve concrete attacks against the "human approval" step that agentic AI systems use before sending money, deleting files or emailing outsiders, then tests them against AP2, Microsoft's action-bound approval design, and MCP elicitation. Auditing ten agent runtimes, the authors found none produced a third-party verifiable approval record, none enforced measurable flood limits, and four of eight runtimes with measurable binding allowed the underlying action's arguments to change after approval. Their reference fix, a gateway called ASU that renders approval cards from a canonical effect descriptor and binds them to a phone passkey signature, cut forged-narration attacks from 0.54 to 0.03 success rate and raised LLM-approver attack catch rate from 0.30 to 0.98 in the

    No. 3 of 5arXiv cs.CRThu 15 Oct
  4. Artificial intelligence

    FDE-Gym: The Last Mile of AI Delivery

    A new benchmark, FDE-Gym, tests AI agents on 95 simulated client-delivery tasks across 18 sectors where, like a real forward-deployed engineer, the agent must clarify ambiguous requirements and discover unstated constraints rather than work from a clean spec. Across 13 model configurations from seven providers, the best setup averaged a normalized score of 0.56 and fully succeeded on only 14 of 95 tasks, with the gap persisting even when the agent had gathered all relevant facts from stakeholders.

    No. 4 of 5Hugging Face BlogWed 14 Oct
  5. Security & privacy

    Language Is a Security Surface for Agent Guards

    Researchers built a benchmark (LinguaGuardBench) that translates only the user instruction in agent-safety test cases into five languages while keeping every tool call, observation and label identical, isolating language as the sole variable in guard verdicts. Across 19 tested guards, most open, small, and purpose-built guards flipped verdicts beyond their own noise floor under translation, with the effect worsening for rarer languages (Amharic worse than French), while frontier guards like Gemini 3.8 Flash stayed essentially unaffected. Smaller guards showed either more false positives or, below 10B parameters, meaningful drops in recall (missed unsafe actions).

    No. 5 of 5arXiv cs.CRThu 15 Oct