RESEARCH
Cryptography and post-quantum surge as hardware attacks test software guarantees
Cryptography dominates the week: 54 papers against an average of 23.2, and post-quantum rises to 26 against 7.5, with much work on implementing standardized schemes. The second thread is hardware getting around software: TPM 2.0, SLH-DSA and LLM inference. AI security falls in numbers but concentrates on agents. Single-week counts remain noisy.
This week's trends
Post-Quantum Cryptography · Rising
26 papers this week · average of the previous 4 weeks: 7.5
Post-quantum rises from an average of 7.5 to 26 papers. Attention is shifting from standards to real systems. PQCD-TLS ties post-quantum signature verification to key derivation, so implementation bugs surface. Another paper measures TLS 1.3 with ML-KEM on microcontrollers. An NTT accelerator also supports Korean schemes, beyond NIST's. Masking against side channels is getting more attention too. For insurers, migration is shifting from algorithm choice to implementation quality: that is what to assess.
Papers cited:
- Verifiable Post-Quantum TLS Handshake: Recognizing Implementation Errors of Cryptographic Verification Steps
- Testing TLS 1.3 PQC on constrained devices and microcontrollers
- A Crypto-Agile Mixed-Radix NTT Accelerator for Negacyclic and Trinomial Rings
- Efficient and Compact High-order Masking Raccoon on Memory Constrained Devices
Cryptography · Rising
54 papers this week · average of the previous 4 weeks: 23.2
At 54 papers against an average of 23.2, classical cryptography is the fastest-growing topic. Beyond work on anonymous credentials and predicate encryption, several analyses question systems many people rely on. One reconstructs the specification of Estonia's IVXV internet voting system and finds new vulnerabilities. An SoK examines design-level backdoors in block ciphers. A study shows physical faults that can cause nonce repetition in threshold lattice signatures. The common message: guarantees rest on assumptions that need checking.
Papers cited:
- Reverse Engineering IVXV: A Specification and Security Analysis of Estonia’s Internet Voting System
- SoK: Design-Level Backdoors in Block Ciphers Constructions, Detection, and Key Recovery
- Fault-Induced Nonce Repetition in Threshold Lattice Signatures: A Reveal-Structure Criterion, Cross-Scheme Analysis, and Detection Boundaries
- Faster and Smaller Anonymous Credentials from Standardized Signatures
- Implementing Predicate Encryption for Branching Programs: Private Computation over Key-Homomorphic Encodings Without FHE
Hardware & Side Channels · Steady
9 papers this week · average of the previous 4 weeks: 7.8
The paper count is in line with the average, 9 against 7.8. Three strong results, however, point the same way: hardware can hollow out software guarantees. TPM 2.0-based full-disk encryption is bypassed on Windows and Linux. A correlation power analysis recovers SLH-DSA's secret seed with SHAKE. A single Rowhammer bit-flip persistently slows LLM inference on GPUs. For insurers, TPM-only disk encryption deserves an extra item in the underwriting questionnaire, for instance on whether a pre-boot PIN is used.
Papers cited:
- Active and Passive Attacks against TPM 2.0-based Full Disk Encryption
- SHAKEing SLH-DSA with Non-profiled SCAs
- Amplifying Latency and Cost of LLM Inference via Rowhammer Attack
AI & LLM Security · Cooling
92 papers this week · average of the previous 4 weeks: 145.2
AI security papers are below their recent average, 92 against 145.2: in a single week this may be noise. The content, though, is consistent, and attention is moving from models to agents. There are prompt injections that become persistent compromise through long-term memory, and prompt injections hidden in forensic evidence. Other papers offer benchmarks on respecting delegated authorization, red teaming of multi-agent systems and in-kernel control of outbound connections. For insurers, agents with memory and broad permissions are a new loss scenario to cover in AI-use questions.
Papers cited:
- Combining Long-Term Memory and LLMs Comes at a High Cost: From Prompt Injection to Persistent Compromise
- Prompt Injection Through Forensic Evidence: A Taxonomy and Proof of Concept of Anti-Forensic Attacks on LLM-Assisted Digital Investigations
- Act on Your Behalf, Beyond Your Authorization: Benchmarking LLM Agent Authorization Alignment
- Black-Box Red-Teaming of Multi-Agent Systems via Reinforcement Learning
- Connection Escrow: Holding AI-Agent Pod Egress in the Kernel Until an Isolated Verifier Rules
How this digest is made
New papers are collected every day and selected every week from arXiv, the IACR ePrint archive and the main security conferences and journals (IEEE S&P, USENIX Security, ACM CCS, NDSS, CRYPTO, PoPETs, IEEE TIFS and others). A score based on source, topic and early signals picks the candidates; an AI model assesses them and writes the summaries. Summaries are based on the abstracts: the original paper is the reference for details.
New papers assessed for this edition: 344.