Post-Quantum Crypto
France sets 2027 deadline: no certification for products without quantum-resistant encryption
France's national cybersecurity agency ANSSI announced it will cease certifying security products that do not include quantum-resistant encryption beginning in 2027. The policy, disclosed at the France Quantum conference and reported by Reuters, effectively compels government agencies and operators of critical infrastructure to migrate away from classical cryptographic systems on a hard timeline — one of the first national-level regulatory mandates of its kind.
Quantum Computing
Provable exponential quantum speedup for topological data analysis demonstrated
Researchers publishing in PRX Quantum have proposed a quantum algorithm that achieves exponential speedup for computing persistence, a core problem in topological data analysis. The result is notable because TDA has real-world applications in biology, materials science, and data science — and provable (not heuristic) quantum speedups for practically motivated problems remain rare. The authors describe this as potentially driving new practical applications in quantum computing.
China's Origin Wukong quantum computer integrates PQC security layer after one million tasks
China's third-generation superconducting quantum computer Origin Wukong, developed by the Anhui quantum computing labs in Hefei, has completed more than one million global computing tasks. The system has now integrated a built-in post-quantum cryptography architecture, making it notable as a vendor claim of hardened quantum hardware security — though independent corroboration of the technical implementation is absent from the announcement.
Louvre routing scheme cuts hardware connectivity demands for quantum LDPC error correction
Published in npj Quantum Information, the Louvre scheme reduces the physical connectivity requirements of quantum low-density parity-check error correcting codes by introducing a routing layer with an expanded quantum instruction set. If confirmed by independent implementation, this could lower the hardware bar for fault-tolerant quantum computing based on LDPC codes.
Quantum Comms
MIT framework generates scalable cluster states from heterogeneous spin emitters with constant overhead
A team led by Dirk Englund at MIT has published in PRX Quantum a control framework for driven inhomogeneous solid-state spin emitters that enables high-fidelity global operations and scalable optically heralded entanglement generation. The key advance is achieving O(1) experimental overhead as the system scales from 10 to 100 qubits — a practical milestone for photonic quantum networking with imperfect emitters.
QKD
Multi-user CV-QKD access network field trial hits Mbit/s key rates over commercial fiber
Researchers demonstrate a downstream broadcast continuous-variable quantum access network (CV-QAN) connecting a Quantum Line Terminal to multiple Quantum Network Units over commercial fiber in a 1:16 field trial. The system achieves Mbit/s-level asymptotic secure key rates and resolves the modulation variance selection problem under link asymmetry using a multi-user utility model — a step toward practical Quantum-to-the-Home deployment.
AI & ML
Nature paper evaluates conversational AI for disease management
A study published in Nature investigates the use of conversational AI for disease management — a significant venue signal for a topic that has typically appeared in lower-tier or preprint channels. The paper's full abstract is not surfaced in the feed, but publication in Nature indicates a rigorous evaluation of clinical dialogue AI with findings the editors judged broadly relevant.
Attention heads in LLMs approximated by Python programs, enabling symbolic inspection
Researchers propose generating executable Python programs that approximate the attention patterns of transformer heads, achieving over 75% Intersection-over-Union similarity on TinyStories. Replacing 25% of attention heads across GPT-2, TinyLlama-1.1B, and Llama-3B with these programmatic surrogates incurs only a 16% average perplexity increase while preserving downstream QA performance — a concrete step toward symbolic transparency in neural models.
CEO-Bench reveals frontier LLMs still fail at long-horizon adaptive business strategy
CEO-Bench evaluates agents on a 500-day startup simulation requiring pricing, marketing, budgeting, and strategic coordination. Only Claude Opus 4.8 and GPT-5.5 finish above the $1M starting balance, and neither reliably turns a profit. The benchmark highlights a persistent gap between short-horizon task execution (where LLMs excel) and sustained adaptive strategy over time.
Google DeepMind and UK government build AI tool to speed housing planning approvals
Google DeepMind announced a partnership with the UK government to develop an AI-powered prototype aimed at accelerating housing planning decisions. The announcement is light on technical specifics and appears to be an early-stage government engagement rather than a deployed system, but signals DeepMind's push into AI for public sector administrative bottlenecks.
Robotics
Qwen-RobotManip tops RoboChallenge, beating π0.5 across OOD manipulation benchmarks
Alibaba's Qwen team reports Qwen-RobotManip, a Vision-Language-Action foundation model built on Qwen-VL and trained on a ~38,100-hour corpus assembled entirely from open-source datasets and human videos. The model introduces a unified alignment framework across representation, motion, and behavior dimensions to enable coherent multi-source training at scale. It substantially outperforms prior state-of-the-art including π0.5 across six out-of-distribution evaluation settings and achieves a 20% relative improvement to rank first in RoboChallenge, with real-robot validation on AgileX ALOHA, Franka, UR, and ARX platforms.