USENIX Security '25 (34^th USENIX Security Symposium)#

Program source is the expanded technical-sessions schedule (titles, authors, and full abstracts; no PDF/slide/recording links were present in the pasted program text, so no Links field is included below — every write-up is grounded in the abstract only). USENIX Security is enormous (380+ accepted papers plus Enigma industry talks across three days); this is a tight cut toward the PM's stated interests — physical DC/WAN networking, cloud-native networking, network security, and identity/authentication — with most of the program (ML/AI privacy, blockchain internals, mobile-app privacy measurement, hardware fault injection unrelated to networking) left out.

Core#

Trust Agility in an Era of Transient Ties: Rethinking TLS Trust in a Rapidly Evolving Ecosystem#

Recording

Hoss Shafagh, Netflix (Enigma track) — Core

• Why: A practitioner-level look at the most consequential near-term shift in the Web PKI — root programs shortening server certificate lifetimes to a maximum of 47 days by 2029 — from someone who runs PKI and workload-identity infrastructure at Netflix scale, not a standards document.
• What: An argument for "trust agility" — the ability to rapidly and securely update trust relationships as certificate authorities, cryptographic standards, and threat models evolve — as the operational response to shortening cert lifetimes, slower-upgrading consumer devices, and growing non-browser (machine-to-machine, workload identity) use cases that fall outside traditional browser trust models. (from abstract only)
• How: Draws on real-world PKI operations at Netflix to describe automation and lifecycle-aware certificate management strategies across diverse endpoints, including preparing for post-quantum cryptographic transitions. (from abstract only)
• Where applicable: Anyone operating PKI/TLS infrastructure at scale, workload-identity systems, or IoT/embedded fleets with slow upgrade cycles; general but grounded in a large-scale production environment.

BGP Vortex: Update Message Floods Can Create Internet Instabilities#

Recording

Felix Stöger, ETH Zurich; Henry Birge-Lee, Princeton University; Giacomo Giuliari, Mysten Labs; Jordi Subira-Nieto and Adrian Perrig, ETH Zurich — Core

• Why: A new BGP instability attack that neither BGPSEC nor RPKI can prevent, because every message involved is a legitimate, standards-compliant BGP UPDATE — a reminder that route-origin validation alone doesn't close the routing-security gap.
• What: Discovery of the "BGP Vortex," a configuration where just three legitimate BGP UPDATE messages trigger persistent instability, weaponizable to cause router overload and forwarding loops across the Internet. All major router implementations tested were susceptible. (from abstract only)
• How: Exploits standards-compliant BGP Communities extensions (used for traffic engineering) that allow route-preference modification; the paper proposes a framework for classifying which BGP extensions are similarly dangerous versus safe to deploy. (from abstract only)
• Where applicable: General for any BGP-speaking network; the mitigation framework is for router vendors and operators using BGP Communities-based traffic engineering.

Ares: Comprehensive Path Hijacking Detection via Routing Tree#

Recording

Yinxiang Tao et al., Tsinghua University / Zhongguancun Laboratory / Quan Cheng Laboratory; Congcong Miao, Tencent — Core

• Why: Path hijacking (manipulating AS-path attributes rather than just the origin) is the stealthier successor to origin hijacking specifically designed to evade existing detection — this is a system that closes that gap with real-event validation, not just simulation.
• What: Ares, a system detecting path hijacking by tracking changes to an AS's "observed prefix routing tree" (OPRT), validated against 12 real-world hijacking events (detected within 5 minutes each) and large-scale simulation (97.2%–99.3% detection rate, 1.06% false-positive rate). (from abstract only)
• How: Uses weighted edit distance between routing trees to quantify anomalous change, combined with clustering to accelerate detection and heuristics to reduce false positives; generates ~2.31 alerts/hour across the entire Internet. (from abstract only)
• Where applicable: General BGP route-hijacking detection for network operators and Tier-1/content ASes; a deployable detection system rather than a protocol change.

Universal Cross-app Attacks: Exploiting and Securing OAuth 2.0 in Integration Platforms#

Recording

Kaixuan Luo et al., The Chinese University of Hong Kong; Samsung Research America — Core

• Why: A platform-wide OAuth account-linking design flaw affecting Microsoft, Google, and Amazon integration platforms — a concrete, high-severity (CVSS 9.6) example of exactly the kind of OAuth/identity-integration failure that matters at the architecture level, not the implementation level.
• What: Two new attacks — Cross-app OAuth Account Takeover (COAT) and Request Forgery (CORF) — exploiting the lack of app differentiation in multi-app OAuth account-linking on integration platforms (Workflow Automation, Virtual Assistants, Smart Homes). 11 of 18 tested platforms were vulnerable to COAT and 5 to CORF. (from abstract only)
• How: Built COVScan, a semi-automated black-box tool that profiles OAuth designs across platforms to find cross-app vulnerabilities; a victim risks account takeover simply by linking any single malicious app, or even clicking a crafted link. (from abstract only)
• Where applicable: General for any platform supporting third-party app marketplaces with OAuth-based account linking; the architectural lesson (differentiate apps, not just users, in the OAuth trust model) generalizes beyond the specific platforms tested.

X.509DoS: Exploiting and Detecting Denial-of-Service Vulnerabilities in Cryptographic Libraries Using Crafted X.509 Certificates#

Recording

Bing Shi et al., Alibaba Group; Luyi Xing, Indiana University Bloomington — Core

• Why: A systematic study of an underexplored class of cryptographic-library vulnerability — availability, not confidentiality/integrity — finding 18 new and 12 previously known CVEs across seven mainstream crypto libraries, all reachable via crafted certificates rather than exotic inputs.
• What: X.509DoS, a class of denial-of-service attacks against cryptographic library implementations, launched via specially crafted X.509 certificates; a tool for rapidly generating such certificates and detecting the underlying DoS vulnerabilities. (from abstract only)
• How: Systematically probes implementations handling X.509 certificates for availability-impacting bugs, showing that spec/standard compliance alone does not guarantee resistance to resource-exhaustion attacks. (from abstract only)
• Where applicable: General for any service parsing/validating X.509 certificates from untrusted input (TLS servers, mTLS endpoints); the finding that standards-compliance ≠ DoS-safety generalizes beyond the seven libraries tested.

Adjacent#

Assessing the Aftermath: The Effects of a Global Takedown Against DDoS-for-hire Services#

Recording

Anh V. Vu et al., University of Cambridge; University of Edinburgh; University of Strathclyde; University of Illinois Chicago — Adjacent

• Why: A rare quantitative, multi-year measurement of whether law-enforcement disruption actually works against the DDoS-for-hire ecosystem — useful context for anyone evaluating DDoS threat trends rather than just technical mitigation.
• What: An assessment of the largest booter takedown effort to date (since December 2022), finding over half of seized sites returned within a day, traffic to re-emerged domains dropped 80–90%, and the first takedown wave cut global DDoS volume 20–40% (with a significant effect specifically on UDP-based attacks) — but the overall market proved resilient, with effects lasting at most ~6 weeks. (from abstract only)
• How: Combines web traffic data, ground-truth visits to seized sites, millions of DDoS attack records, and underground-forum/Telegram chat analysis. (from abstract only)
• Where applicable: General threat-intelligence context for DDoS trend forecasting; policy-relevant more than technically actionable for network defenders directly.

An Industry Interview Study of Software Signing for Supply Chain Security#

Recording

Kelechi G. Kalu et al., Purdue University — Adjacent

• Why: Software signing is widely recommended by supply-chain security frameworks but adoption remains low — this is the first in-depth practitioner account of why, complementing the technical PKI/cert-validation picks above with the human/organizational side.
• What: Interviews with 18 experienced security practitioners across 13 organizations on software-signing adoption, surfacing technical, organizational, and human barriers; finds experts disagree on signing's importance and documents how internal/external events (breaches, new regulation) drive adoption. (from abstract only)
• How: Qualitative interview study producing a refined software-supply-chain factory model highlighting where and why signing practices break down. (from abstract only)
• Where applicable: General for organizations evaluating or mandating software signing as part of supply-chain security policy.

Wildcard#

Catch-22: Uncovering Compromised Hosts Using SSH Public Keys#

Recording

Cristian Munteanu et al., Max Planck Institute for Informatics; Delft University of Technology (Distinguished Paper Award Winner) — Wildcard

• Why: A clever, minimally-invasive Internet-scale detection technique — using SSH's own public-key-authentication challenge behavior to fingerprint compromise without ever needing valid credentials or touching the compromised system — that found state-actor-associated keys inside sensitive ASes.
• What: A method identifying 21,700+ compromised SSH servers across 1,649 ASes and 144 countries by checking for the presence of known-malicious attacker public keys, without needing to log in or test passwords; surfaced new context on botnets (fritzfrog), threat actors (teamtnt), and state-actor-linked keys in sensitive networks. (from abstract only)
• How: Exploits the fact that SSH only sends a public-key-authentication challenge if the corresponding key is actually present on the server — letting a scanner test for known malicious keys' presence with zero access and zero risk of triggering an actual login. (from abstract only)
• Where applicable: General Internet-scale compromise detection; the technique's elegance (a side effect of protocol behavior, not a vulnerability) is the interesting part, applicable anywhere an authentication protocol leaks presence information.

Red Bleed: A Pragmatic Near-Infrared Presentation Attack on Facial Biometric Authentication Systems#

Recording

Bowen Hu, Kuo Wang, and Chip Hong Chang, Nanyang Technological University — Wildcard

• Why: A practical, sub-$400 hardware attack defeating a widely-deployed commercial biometric authentication system (Windows Hello, confirmed by Microsoft's own security response center) — a concrete demonstration that "commercial-off-the-shelf NIR biometric = secure" is no longer a safe assumption.
• What: Red Bleed, a presentation attack against near-infrared (NIR) facial biometric authentication (e.g., Windows Hello, Apple Face ID) using a custom-built consumer LCD display costing under $400, converting ordinary visible-light face images/video into convincing NIR spoofs. Confirmed against a commercial Windows Hello module with a CVE scheduled for disclosure. (from abstract only)
• How: A generative framework (VAE-based) converts RGB/visible-spectrum face images or video into the NIR domain, combined with face-swapping, so attackers can use easily-obtainable RGB material (social media photos, video calls) rather than needing rare NIR footage of the target. (from abstract only)
• Where applicable: NIR-based facial biometric authentication specifically (a widely deployed modality); a hardware/software technique that generalizes to any NIR liveness-detection system relying on the assumption that commodity displays can't emit convincing NIR.

Themes#

• BGP/routing security keeps finding fresh, standards-compliant attack surface. BGP Vortex (legitimate UPDATE messages triggering instability) and Ares (stealthy path hijacking bypassing origin-only defenses) both show that RPKI/ROV-class defenses, while necessary, don't close the routing-security gap — the protocol's own legitimate extensibility is the remaining attack surface.
• Tunnel and overlay endpoints are under-defended at Internet scale. "Haunted by Legacy" found 4M+ open GRE/IPIP/4in6/6in4 hosts — the same encapsulation families underlying DC/WAN overlay networking — a concrete reminder that tunnel endpoints need the same perimeter discipline as any other network boundary, not an implicit trust assumption.
• Identity infrastructure is being stress-tested at the integration layer, not just the crypto layer. Passkey cloud-backup compromise (CASPER), OAuth account-linking takedown (Universal Cross-app Attacks), and formally verified X.509 validation (Verdict) all target where standards meet real deployment — the place bugs actually live.
• Crypto-agility has gone from aspiration to forcing function. The Netflix Enigma talk on 47-day certificate lifetimes by 2029 sets the operational stakes; the post-quantum secure-messaging cluster (Signal, iMessage PQ3, MLS) shows the same agility pressure playing out at the protocol-design layer for already-deployed systems, not hypothetical future ones.
• 5G/cellular core and RAN security remains a rich, mostly-unsolved vertical. Multiple independent groups this cycle found fundamental protocol-level gaps (AKMA linkability, pre-auth message injection, core-network fuzzing crashes, geofenced DoS) — worth periodic monitoring even outside a primary cloud-native/DC focus, since the control/data-plane separation patterns echo modern network function design.
• The most memorable findings this cycle are cheap, elegant techniques, not brute-force ones. A sub-$400 LCD spoofing commercial biometric hardware, an SSH protocol side-effect fingerprinting 21,700+ compromised hosts with zero access, and three legitimate BGP messages destabilizing routers all make the same point: novel low-cost techniques against deployed systems remain more interesting than incremental ML-classifier papers.