Black Hat USA / Europe 2024-2025 research roundup

Black Hat USA / Europe 2024-2025 research roundup

TL;DR: Black Hat USA and Europe in 2024-2025 leaned heavily on cloud identity abuse (Entra/Okta cross-tenant sync, OAuth device-code), Windows kernel and EDR bypass tradecraft, browser exploit-chain work, automotive and embedded breaks from the Pwn2Own pipeline, the first wave of “real” AI/LLM agent attacks, and hardware/firmware disasters such as PKfail, LogoFAIL, and the xz-utils supply-chain backdoor. This note is the conference-talk companion to case-study-portswigger-top-10-pattern and pwn2own-2024-2025-research-roundup, and pairs well with keeping-up-with-research-feeds for ongoing intake.

Why it matters

Black Hat is one of the few venues where vendor-disclosed, peer-reviewed offensive research lands with both slides and recorded talks, and where defenders, red teamers, and bug-bounty hunters all consume the same primary source. Watching a season of BH talks is a fast way to:

  • Calibrate which classes of bugs the top researchers think are worth six months of effort.
  • Pick up new techniques before they hit blog posts, training courses, or HackTricks-style aggregators.
  • Spot the gap between “cool demo” and “real impact” so your own reports (see demonstrating-impact and report-writing-for-pentesters) land better.

If you only have time for one ingestion ritual, BH USA briefings + BH EU briefings + Pwn2Own write-ups is the strongest signal-to-noise loop available.

Themes and notable talks

Cloud identity (Entra, Okta, M365)

Identity took over the BH USA Enterprise track in both years. Standout work:

Takeaway: identity is the new perimeter, and BH is publishing the playbook a year before it shows up in red-team engagements.

Windows kernel and EDR bypass

  • Multiple talks on BYOVD (Bring Your Own Vulnerable Driver) taxonomies, including improvements to LOLDrivers and EDRSandblast-style tooling.
  • Connor McGarr and others on Windows kernel exploit primitives post-HVCI and post-VBS, focusing on arbitrary-read/write to kernel-mode code execution under hardware-enforced stack protection.
  • Several sessions on WDAC and AppLocker bypass via signed-binary abuse and policy-merging tricks, useful input for edr-rules-as-code-from-attack-patterns and detection-engineering-pyramid-of-pain.
  • EDR-internals talks dissecting userland hooking, kernel callback enumeration, and ETW-TI evasion. Tie these back to atomic-red-team-emulation-deep for emulation coverage.

Browser exploitation

  • V8 and JavaScriptCore exploitation chains, often referencing Pwn2Own entries that later appeared at BH with more detail. See pwn2own-2024-2025-research-roundup.
  • Renderer-to-sandbox and Site Isolation weakness research, including content-process compromise leading to UXSS-like primitives.
  • Talks on WebGPU, WebAssembly, and WebCodecs as new attack surfaces - all relatively young APIs with weaker fuzzing maturity.
  • Recurring theme: modern browser exploits increasingly need two or three primitives chained, and the “single-bug RCE” era in mainstream browsers is essentially over.

Web research and HTTP-layer attacks

Automotive and Pwn2Own-adjacent

  • Tesla infotainment and modem breaks from the Pwn2Own Automotive Tokyo events were re-presented at BH with deeper technical detail.
  • Lennert Wouters (KU Leuven COSIC) and collaborators continued automotive immobiliser and key-fob cryptographic-implementation research, plus broader embedded fault-injection work that connects to fault-injection-laser-emfi and hardware-glitching-deep.
  • EV charging infrastructure (OCPP, ISO 15118) showed up as a fresh, under-tested attack surface.

AI and LLM security

  • First wave of prompt-injection-as-real-exploit talks: indirect injection via documents, emails, and tool outputs leading to data exfiltration in agentic systems.
  • Model supply-chain talks on Hugging Face artefact abuse, pickle deserialisation, and malicious model weights.
  • LLM red-teaming methodology sessions building on OWASP LLM Top 10 and MITRE ATLAS, pairing well with ai-agent-sandbox-design.
  • Defensive talks on agent sandboxing, capability scoping, and tool-call authorisation.

Honest framing: most 2024 AI talks were demo-heavy and impact-light. 2025 talks started to show real production-system compromises, particularly in enterprise copilot integrations.

Hardware, firmware, and supply chain

  • PKfail (Binarly): leaked Platform Keys shipped by major OEMs allowing Secure Boot bypass at scale. See pkfail-uefi-secureboot-bypass and bootloader-and-secure-boot-attacks.
  • LogoFAIL follow-ups: image-parser vulnerabilities in UEFI firmware, expanding the pre-OS attack surface.
  • xz-utils backdoor (CVE-2024-3094): dedicated retrospectives on the social-engineering long-con, the obfuscated payload, and the detection story. Direct input to case-study-3cx-supply-chain and case-study-solarwinds-2020 as a comparison set.
  • Continued TPM, fTPM, and Secure Enclave research, including sniffing and replay attacks on discrete TPM buses.

Cryptographic implementations

Detection, IR, and threat intel

BH USA’s defensive tracks added value too:

Patterns across the season

A few patterns repeat across both years:

  1. Chains beat single bugs. Almost every headline talk required two or more primitives. Build your own research the same way - assume one bug is not a finish line.
  2. Identity, firmware, and AI are the three growth areas. Web and kernel research is still strong but increasingly incremental.
  3. Tooling release at talk time. Most serious BH talks ship code (BloodHound modules, fuzzers, EDR-evasion frameworks). Track repos, not just slides.
  4. Disclosure tracks matter. Many BH talks coordinate with vendors on a multi-month timeline; the embargo story is itself a learning artefact for disclosure-and-comms and responsible-disclosure-across-jurisdictions.

Workflow to study a BH season

  1. Pull the briefings index for BH USA and BH EU as soon as it is published. Skim every abstract; star 20-30 talks.
  2. Watch the recordings once they hit the official YouTube channel (usually a few weeks post-event). Take notes on technique, not narrative.
  3. Grab the slides and white paper from the briefing page; many talks publish a longer paper than the talk covers.
  4. Find the repo. If the talk released tooling, clone it into building-a-research-home-lab and reproduce one primitive.
  5. Map to your existing notes. For each talk, ask: which of my topic notes does this update? Edit them directly rather than hoarding a “to read” list.
  6. Cross-reference Pwn2Own. Many BH talks are extended versions of Pwn2Own entries. Use pwn2own-2024-2025-research-roundup as a cross-index.
  7. Feed your RSS/Atom intake. Add researcher blogs to keeping-up-with-research-feeds so the next year’s work shows up before the conference.

Defensive baseline

If you are on the blue or purple side, the minimum follow-up from a BH season:

  • Patch firmware (UEFI, TPM, baseband) on a defined cadence, not just OS patches.
  • Review Entra / Okta cross-tenant settings and conditional-access posture against the latest talks.
  • Add detection content for the year’s BYOVD drivers and any new EDR-evasion primitives.
  • Re-baseline your supply-chain controls against xz-style social engineering: maintainer takeover, obfuscated build steps, test-fixture payloads.
  • Update tabletop exercises with the season’s most plausible chains; see purple-team-feedback-loop and ir-from-source-signals.

References

  • Black Hat USA briefings archive: https://www.blackhat.com/us-24/briefings/schedule/ and https://www.blackhat.com/us-25/briefings/schedule/
  • Black Hat Europe briefings archive: https://www.blackhat.com/eu-24/briefings/schedule/ and https://www.blackhat.com/eu-25/briefings/schedule/
  • Binarly PKfail research: https://www.binarly.io/blog/pkfail-untrusted-keys-undermining-secure-boot
  • PortSwigger Research (James Kettle): https://portswigger.net/research
  • Orange Tsai (DEVCORE) blog: https://blog.orange.tw/
  • xz-utils backdoor analysis (Akamai SIRT): https://www.akamai.com/blog/security-research/critical-linux-backdoor-xz-utils-discovered-what-to-know