BSidesSF 2026
2026-03-21..2026-03-22
San Francisco, United States
BSidesSF ·
[BSidesSF 2018, BSidesSF 2019, BSidesSF 2020, BSidesSF 2023, BSidesSF 2024, BSidesSF 2025, BSidesSF 2026]
| title | author | length |
|---|---|---|
| Opening Remarks (Saturday) | Reed Loden | 15m |
Welcome to BSidesSF’s kickoff with the conference framed as security theater and a call to participate not spectate; organizers report 18 program changes so… | ||
| Let's Do the Timewarp Again! A Look Back to Move Forward | Anna Westelius | 41m |
A concise history of what actually changed: encryption becomes default after widespread push and the emergence of Let's Encrypt made HTTPS free and automated… | ||
| The Heist: Chasing an Advanced Crypto Attacker Across the Multi-cloud | Yotam Meitar | 47m |
A multi‑cloud intrusion steals about $107M over seven months, culminating in an $80M heist; attackers social‑engineer the help desk to reset an approver’s… | ||
| 📕 AI-Powered AppSec: 10x Your Security Team Without Scaling Headcount 🖥️ | Anshuman Bhartiya | 45m |
Built an internal platform called Priscia that runs CI hooks into a Flask backend, stores artifacts in DynamoDB, and queries multiple LLMs (Sonnet/Opus and… | ||
| Your Load Balancer is Your New Perimeter: Attacks & Defenses at Scale | Arjun Sharma | 32m |
Four practical bypasses show how origin IPs leak and let attackers bypass CDNs: a certificate transparency leak reveals subdomains that resolve to load… | ||
| The Phantoms of the Fraudpera: An Overview of Anti-Detection Tooling | Bobbie Chen | 33m |
Overview of commercial evasion tooling and attacker economics: four concrete categories—residential proxies, CAPTCHA solvers, anti-detect browsers, device… | ||
| Detection at Scale: Abstracting Detection Intent | Gaurav Singh, Dario Amiri | 45m |
Claim: abstracting detection intent lets large teams and high event volumes scale detections by separating declarative pattern expression from system… | ||
| State of (Absolute) AppSec | Seth Law, Ken Johnson, Kevin McDermott, Astha Singhal, Clint Gibler | 40m |
Panel argues application security must adapt to non-deterministic AI-infused apps where models change runtime behavior, not just scan outputs. Concrete changes… | ||
| How to Be a GRC Hero (Without Heroics) | Stas Bojoukha | 38m |
Argues that spreadsheets and fire drills are obsolete and shows a working playbook: connect telemetry streams to controls so an MFA drop 99%→95% flips control… | ||
| 📙 Is Q-Day Worse than Y2K? Strategies for Surviving the Quantum Threats | Sandip Dholakia | 43m |
Claims that Q‑Day arrives in 5–10 years rest on steady hardware progress and a surprising estimate that breaking 256‑bit ECC needs ~317 million qubits (hour)… | ||
| Follow the data to learn the secret | Dylan Ayrey | 35m |
Scans of HuggingFace (25% so far) with Truffle Hog process millions of files across ~880,000 datasets and flag almost 100,000 live secrets, including API keys… | ||
| 📗 Orchestrating Resilience: Composing a New Score for Netflix Service Reliability 🏗️ | Sandhya Narayan, Prachi Jain | 28m |
Netflix replaces blanket freezes with a service tiered service model and a data-driven risk-aware launch rubric wired into CI/CD pipelines, classifying… | ||
| Securing Space: The Next Frontier for Security Engineers | Anshu Gupta | 46m |
Satellites look like code plus hardware and the talk shows concrete attacks, tooling, and defenses: the KSAT/Viasat-style outage via firmware overwrite and the… | ||
| From $10 to $30M: Operating in the Data-Extortion Aftermath | Diego Matos | 35m |
Claims that multi-level extortion now mixes encryption, public leakage, DDoS and customer harassment, with active groups up 49% and leak events up 50%, and… | ||
| Practice Cyber Skills Like a Musician | Bianca Ionescu | 31m |
Argues that musical discipline maps to cybersecurity through deliberate practice: drills, slow breakdowns, and repetition to build muscle memory for commands… | ||
| The Phaaaaaaaaantom of the Salt Typhoon is there, inside i-SOON | Daniel Schwalbe | 41m |
Claims a reproducible opening in Chinese-sponsored APT Salt Typhoon’s TTPs by mapping domain infrastructure from a 2024 GitHub leak, using passive DNS pivots… | ||
| Reverse Engineering Go Malware: From Manual to AI-Powered Analysis | Asher Davila | 47m |
Shows step-by-step reverse engineering of Go IoT malware (BotenaGo, PumaBot) using Ghidra, Radare2, r2ai and GhidraMCP to rebuild strings and payloads and map… | ||
| 📙 Your Threat Model Is Lying to You: Why Modeling the Design Isn’t Enough in 2026 🖥️ | Farshad Abasi | 47m |
Claim: threat models stop at design and miss deployed reality, so production scans must feed them. Gives numbers (94% of cloud infra has manual pieces, 6%… | ||
| 📗 Gettings PCAPs from Stingrays for $20 with Rayhunter | Cooper Quintin, Will Greenberg | 46m |
A $20 Android Orbit hotspot running Qualcomm DIAG becomes Rayhunter, a Rust tool that dumps QMDL/pcap frames and applies heuristics (2G downgrade, null cipher… | ||
| 📙 AI as an Accountable Entity: Governing Risk When Machines Make Decisions ⚖️ | Pavithra Pradip | 27m |
AI decisions become risk-bearing services that need inventory, a simple materiality score (impact, autonomy, sensitivity) and assigned owners across model… | ||
| Lessons Learned from Building Custom Hacker Hardware | c4m0ufl4g3 | 45m |
Build story of Injectal and Hide: starts with off‑the‑shelf parts—Trinket M0, Arduino Nano 33 IoT, SparkFun XB radio and a DigiMesh radio—total prototype cost… | ||
| AI for security - friend or foe? | Tom Alcock, Jackie Bow, Travis McPeak, Drew Hintz, Kyle Polley | 42m |
Panelists claim AI already cuts alert backlog to zero at one company and drops mean time to triage to minutes, describes an open‑sourced “easy agents” cloud… | ||
| So you think you can airgap? (No.) | Ziyad Edher | 33m |
Anthropic deploys and operates an egress limiter that literally counts every byte leaving training clusters, implements a large-burst token bucket to allow… | ||
| Detecting Race Conditions on macOS | Olivia Gallucci | 34m |
GCD misconfiguration causes real exploitable timing windows: a missing target queue on an XPC listener lets handlers run concurrently and led to GSSCred CVE… | ||
| Anatomy and Defense of LOTL Fileless Intrusions | Amol Sarwate | 44m |
84% of incidents in 2025 involve living-off-the-land binaries and 47% of initial access is via click‑to‑fix lures, the talk demonstrates. Asteroth uses LNK →… | ||
| Not My Vibe: When AI Coding Agents Go Off the Rails | Aonan Guan, Zhengyu Liu | 46m |
Reversing Gemini CLI, Claude Code and Codex uncovers concrete attack paths: command injection via shell-tool gaps, workspace escapes through symlink and… | ||
| 📗 Rehearsal is Over: Moving GRC Engineering from Theory into Practice ⚖️ | Branden Rosenlieb | 31m |
Argues GRC must become an engineering function and shows a working pipeline: AWS audit playbook (Python + boto3) pulls configs to JSON, Terraform plan→JSON… | ||
| 📙 Level Up Your Threat Modeling: Turning Security Into a Team Adventure 🖥️ | Stanley Harris | 31m |
A method that gamifies threat modeling into a role‑playing session claims to make developers stick with the process by mapping real systems, playing from a… | ||
| The Trusted Platform Module | Eric Chiang | 29m |
Explains that the TPM is an API spec spanning thousands of pages implemented as discrete chips or fTPM on CPU, and shows live interaction using a Go package… | ||
| 📙 What a False Alarm Taught Us About Security as a 2-Person Startup 🔥 | Alex Chantavy, Kunaal Sikka | 30m |
Two-person startup with $4.2M seed and first enterprise customers migrates from EC2 to ECS Fargate and a hosted Neo4j Aura graph; a failed local login… | ||
| Elevating First-Time Female Voices on Stage | Poorna Rajaraman, Deepika Gupta | 31m |
A company-run six-month mentorship engine pairs novice-only applicants with experienced mentors matched by domain and time zone, uses a double-blind Technical… | ||
| CISO Series Live Podcast Recording | David Spark, Mike Johnson, Sara Madden | 49m |
Live podcast records a 45-minute episode that airs May 5 featuring rapid audience games, vendor-selection rules like no three-year contracts up front… | ||
| 📙 Prompt, Commit, Repeat: Security at Scale When 1,000 Devs Go AI-Native | Balachandra Shanabhag | 45m |
Enterprises with thousands of developers face concrete failures: an AI agent nuked a Mac by deleting a home directory, prompt and code injections cause… | ||
| 📗 Threat Chords: Tuning into Persistent Patterns in Adversary Behavior 🔥 | Karthika, Samhita Vempatti | 44m |
Claim: attackers reuse operational playbooks so defenders should detect sequences not single IOCs. Evidence: Scattered Spider/Shiny Hunters repeatedly use… | ||
| 📗 Cringe, Corrected: Hot Takes Fixed by the CIS Controls ⚖️ | Lawrence Cruciana, Amelia Cruciana | 26m |
Session mines real social posts and maps bad advice to concrete CIS actions: identifies CentOS 6 is EOL and explains why unpatched OSes defeat FIDO2/SSH keys… | ||
| 📙 From Assistant to Assassin: Weaponizing An OpenClaw Vulnerability to Achieve 1-Click RCE 🖥️ | Mav Levin | 31m |
OpenClaw contains a configuration-logic flaw that lets a crafted URL set a gateway, trigger client-side auto-login, and exfiltrate the admin authentication… | ||
| 📗 You Just Might Find, You Get What You Need: How MS Became My Unlikely Success Story | Emily Harden | 25m |
Rapidly diagnosed with MS after five MRIs and two spinal taps and a five‑day hospital stay, the story shows concrete pivots: three months on FMLA, trading… | ||
| 📙 The Room Where It Happens (Identity Compromise Edition): Behind the scenes of Okta attack campaigns 🔥 | Julie Agnes Sparks | 21m |
Okta shows 1,081 event types but useful hunting focuses on authentication and policy evaluation events; look for sessionId and dthash fields plus debugContext… | ||
| Increasing the Analysis Surface of Large Language Models | Stephen Brennan, Ulrich | 46m |
Claim: model internals reveal attack signals that black‑box filters miss. The team develops attention-based analysis linking unsafe prompts to attention… | ||
| Architecting the Modern SOC: The Evolving AI Reality for Blue Teams | Bryan Fite, Dean De Beer, Nicole Grinstead, Swathi Joshi | 45m |
Architectural choices for a modern SOC determine whether AI assistance helps or breaks operations: prioritize structured data and a federated data fabric… | ||
| Demystifying File Similarity for Malware Detection | Udbhav Prasad | 40m |
Compares fuzzy hashes and ML on 800k labeled PE features from the EMBER dataset and AVclass labels from CrowdStrike, evaluating similarity with label… | ||
| Practical (and impractical) git commit signing | Matthew Garrett | 30m |
Argues that commit signatures can give provenance but real tooling breaks the assurances; demonstrates working code and CI checks (links and repo provided) and… | ||
| 📗 Your Arch-Nemesis is a Data Scientist: What's the Difference Between Security and Privacy Work? ⚖️ | Aleatha Parker-Wood | 30m |
Privacy focuses on purpose and use not just confidentiality, and the real adversary is often data scientists as adversary who reidentify or retain data… | ||
| When the supply chain hits a sour note | Kennedy Toomey | 25m |
Chronicles an npm compromise that led to malicious releases downloaded 2.5 million downloads with only ~$500 stolen, then a Singularity GitHub Actions… | ||
| 📗 Web standard consortiums are a game with Chrome as the monopoly man | Simon Wijckmans | 32m |
Argues that web standards stall because most browsers run on Chromium dominance so a single vendor effectively vetoes progress, cites 40% Google presence at… | ||
| More Role Models in AppSec: How to Get It Right | Alexandra Charikova | 22m |
After 100+ interviews the claim is clear: AppSec fails more from human gaps than tooling, and visibility fixes work better than more scanners. Data points… | ||
| Composing the Response: Building an Incident Pipeline from Scratch | Geet Pradhan | 29m |
Builds a zero‑budget incident pipeline for early‑stage teams (employee #1/#2) that trades improvisation for plumbing: single intake, role-based command… | ||
| 📙 Conducting the Kill-Chain: Detecting APT Progression Through Music-Sequence Modeling 🤖 | Krupa Brahmkstri, Sneha Rangari | 31m |
Sequence-aware detection engine uses music-inspired encoding, transformer attention and motive detection to score event progressions as harmonic dissonance… | ||
| Breaking Tokens: Modern Attacks on OAuth, OIDC, and JWT Auth Flows | Bhaumik Shah | 33m |
Shows a 2025 Salesforce OAuth breach where a single integration token lets attackers access 7,000+ customer environments, reproduces token replay against a… | ||
| "Ask the EFF" Panel | David Greene, Katharine Trendacosta, Samantha Baldwin, Cooper Quintin | 46m |
EFF panel lists concrete work and wins: about 36 years of org history and roughly 14 active litigations; a threat lab that finds malware on low-end kids’… | ||
| 📙 Incident Readiness You and Your Leaders Will Actually Trust 🔥 | Shachar Hirshberg, Hadar Waldman | 29m |
A repeatable process called an environmental intelligence report queries existing logs (Athena, Splunk, S3), stores queries in notebooks, and uses LLM skills… | ||
| 📙 RBAC Atlas: Mapping Real-World Kubernetes Permissions and Exposing Risky Projects 🏗️ | Lenin Alevski | 35m |
A tool called RBAC Atlas maps over‑permissive Kubernetes identities by running a CLI policy analyzer against Helm charts and YAML and publishing daily results… | ||
| Building an open source security... (Fletcher Heisler, Marcelo Elizeche Landó) | Fletcher Heisler, Marcelo Elizeche Landó | 28m |
Open source IDP Authentic reaches 1M+ installations and scales from a Django codebase to a mixed stack with a partial Go rewrite and an active Rust rewrite PR… | ||
| How We Red-Teamed Our Own AI Agent: Lessons from... (Josiah Peedikayil, HS) | Josiah Peedikayil, HS | 37m |
Operation Pale Fire red-teams Goose, Block’s open-source AI agent and demonstrates an end-to-end chain: sending external invites via Google Calendar MCP… | ||
| Detection Allegro: Composing Detection Rules with... (Raphael Ruban, Chen Cao) | Raphael Ruban, Chen Cao | 30m |
Claims an 80% reduction in detection development and maintenance time by wiring a prompt-builder plus a coding agent into the CI: Gather log schemas and… | ||
| You’re Gonna Be Popular: Why They’re Getting a... (Ruby Murphy, Clea Ostendorf) | Ruby Murphy, Clea Ostendorf | 32m |
Hiring isn’t personal but it is noisy: four root causes—AI resumes vs AI screeners, volume-driven job descriptions, higher automated expectations, and lack of… | ||
| A Worm in the Apple: Wormable Zero-Click RCE in AirPlay... (Avi Lumelsky, Uri Katz) | Avi Lumelsky, Uri Katz | 41m |
Researchers report 23 AirPlay flaws that chain into a wormable zero-click RCE and demo full exploits on a MacBook, a Bose Wi‑Fi speaker and a Pioneer CarPlay… | ||
| The great SAST dissonance: how to please every... (Claudio Merloni, Romain Gaucher) | Claudio Merloni, Romain Gaucher | 51m |
SAST coverage misses the long tail: analysis of ~80,000 repositories surfaces ~7,000 unique dependencies and shows the top 1,000 libraries fully cover only… | ||
| A blueprint for building a generic... (Ashwin Sidhalinganahalli, Fletcher Ramee) | Ashwin Sidhalinganahalli, Fletcher Ramee | 42m |
A hub‑and‑spoke control plane called Guard stores all policies and pushes them to Topaz sidecar evaluators that run next to apps for millisecond P99 latency… | ||
| Kidnapping a Library: How Ransomware Taught the British Library to... (Brian Myers) | Brian Myers | 41m |
Ransomware group Rita compromises valid credentials via terminal services because a critical domain lacked MFA exception, stays undetected for days, then… | ||
| Breaking Endpoint Anti-Ransomware: Going... (Nishant Sharma, Vivek Ramachandran) | Nishant Sharma, Vivek Ramachandran | 35m |
Claims that ransomware can operate entirely in the browser and bypass EDR/SASE/SSE/DLP by abusing web-native controls, demonstrated with an OAuth consent… | ||
| Hunting Malicious IDE Extensions: Building Detection at Scale... (Vinod Tiwari) | Vinod Tiwari | 29m |
Developer workstations run IDE extensions with full user privileges across VS Code Cursor and JetBrains so a detection pipeline is built that inventories… | ||
| Making WAF Mainstream: From Static Defenses to... (Roy Weisfeld, Surya Pentakota) | Roy Weisfeld, Surya Pentakota | 43m |
Shows a working edge security platform that layers AI over CDNs and WAFs to auto-enrich logs, backtest rules in log-only mode, and generate targeted… | ||
| From Auditions to Opening Night: Selecting Security Tools that... (Saurabh Sharma) | Saurabh Sharma | 39m |
Shows a repeatable selection process: baseline finds ~6,000 vulnerabilities with 13 critical fixes a year and a 325‑day mean time to remediate; pick vendor… | ||
| From pocket to Pwn: How we hacked a multinational Corp for $200 with... (Tim Shipp) | Tim Shipp | 29m |
Claims a full-domain compromise using off‑the‑shelf gear for about $200 (real cost ~$125 after the $25 Android developer fee). Uses a magnetic nRF24 card… | ||
| Red Teaming from outside: Identifying and exploiting SaaS systems... (Rojan Rijal) | Rojan Rijal | 39m |
Shows concrete attack chains: abuse of Amazon SES mail-from subdomain to deliver phishing that bypasses SPF/DKIM checks; exploiting unauthenticated GETs used… | ||
| From Noise to Notes: Orchestrating SAST with Developers... (Adrián Puente Z.) | Adrián Puente Z. | 31m |
Walkthrough shows how a SAST rollout uses Snyk Pro and Semgrep to cut 5,000 noisy findings to ~800, prioritize by business risk and scan 40% of repos that… | ||
| Security for AI Agents Using an Ensemble of Fine-tuned... (Lidan Hazout, Bar Kaduri) | Lidan Hazout, Bar Kaduri | 38m |
Runtime guardrail intercepts agent tool calls pre-invocation interception and runs an ensemble of fine-tuned SLMs (4–12B) in parallel — instruction‑violation… | ||
| Power Dynamics in Security Leadership: a legato leitmotif... (Sarai Rosenberg) | Sarai Rosenberg | 28m |
Claims that authoritative directives always withdraw trust and that trust is faster to lose than to build; a four-step method to influence without… | ||
| Closing Remarks (Reed Loden) | Reed Loden | 12m |
Record attendance nears 3,000 total badges, with 92 sessions (2 keynotes, 86 talks, 4 panels), 602 CFP submissions and 154 presenters producing 3,766 slides… | ||
| Saving Bug Bounties from AI Slop | Anto Joseph | 32m |
Bug bounty platforms drown in AI-generated spam and nonreproducible reports, so the talk demonstrates a working system that uses ZK-TLS/TLSNotary MPC… | ||
| Opening Remarks (Sunday) | Reed Loden | 15m |
Welcome back to Day Two with specifics: more than 2,000 people attended yesterday, 58 talks ran across 16 theaters, and the CTF shows an eight-way tie as of… | ||
| Against the Tyranny of Optimization: On the Stability of Automated Republics | Katie Moussouris | 40m |
Against the Tyranny of Optimization: On the Stability of Automated Republics
Katie Moussouris
AI is transforming cybersecurity faster than institutions can… | ||
| 📙 The Great Credential Caper: How to Perform and then Defend Against the (Nearly Impossible) to Defend 🏗️ | Christo Roberts, Dan Hollinger | 41m |
Claim: 41% of logins match previously exposed credentials and during a Black Friday week 95% automated during Black Friday week, attackers use Claude Code with… | ||
| We Pwn the Night: Growing & Leading an 31337 security research team | Keith Hoodlet | 46m |
Goal: hire >25% headcount and have every new researcher find zero days in 45 days. Process pairs attitude, aptitude, engagement screening with a C/C++ or Go… | ||
| Who Watches the NPM Watchers? | Paul McCarty | 38m |
Research uses canary packages — NPM packages embedding Think Canary tokens (JavaScript payloads, EXEs, PDFs, .env and AWS credential decoys, images in READMEs)… | ||
| What happened to the lock icon? | Serena Chen | 26m |
Chrome removes the familiar lock icon in a 2023 redesign after evidence the icon misleads users: a 2021 study finds only 11% correctly identified what the lock… | ||
| 📕 CloudShell Hide-n-Seek: enjoying the sweet persistent sounds of silence! 🔥 | Jenko Hwong, Chris Ryan | 46m |
A multi‑cloud CTF shows an attacker chaining an SSRF to IMDS to dump temporary API creds, enumerate S3, then use those creds to fetch a federated console URL… | ||
| The AppSec Poverty Line: Minimal Viable Security (Tanya Janca) | Tanya Janca | 41m |
Defines an “AppSec Poverty Line” as the smallest set of controls under‑resourced teams must implement: input validation, parameterized queries, output… | ||
| Pwning and Defending AI Agent Code Interpreters (Kinnaird McQuade) | Kinnaird McQuade | 45m |
Claims live exploitability of code interpreters: prompt-to-RCE, credential theft, and covert C2 payloads. Uses interact and a bespoke DNS C2 that encodes… | ||
| The Epistemology of Trust (Mike Wilkes) | Mike Wilkes | 34m |
Argues for measuring resilience by breach cadence (how often you get hit) instead of probability and shows concrete fixes: use EPSS v4 to prioritize reachable… | ||
| Sandboxes, Seccomp, and Syscalls: Chasing Isolation in Kubernetes (Mark Manning) | Mark Manning | 43m |
Covers practical limits of seccomp for Kubernetes: tracing tools (strace, Inspector Gadget, tracie) give inconsistent syscall sets across the same image, a… | ||
| Security on a Shoestring: A Low-Budget Security Revival Tour (Jared Casner) | Jared Casner | 33m |
Argues that a credible security program is possible on a tiny budget by prioritizing basics: passkeys and phishing-resistant MFA, SPF/DKIM/DMARC, short‑lived… | ||
| The Risky Business of Risk Illiteracy (Sean Juroviesky) | Sean Juroviesky | 26m |
Claim: 68% of breaches stem from non‑malicious human factors and 98% of those would be stopped by enforcing MFA. Build an inventory, layered data‑flow diagrams… | ||
| Running an efficient bug bounty program and PSIRT... (Garrett McNamara, Jeff Guerra) | Garrett McNamara, Jeff Guerra | 45m |
Covers concrete steps: implement a two-way vendor API integration, add Slack chat‑ops or Tines automations to auto‑triage and escalate, build a ServiceNow app… | ||
| One Thousand and One AI-Prevented CVEs: Vibe Coding a Whole New... (Brandon Wu) | Brandon Wu | 44m |
Pipeline combines a deterministic callgraph built with Semgrep static analysis and GitHub usage scraping with selective LLM research to auto‑generate tests and… | ||
| How the Vietnam War created single sign-on — and how it's evolved... (Connor Peshek) | Connor Peshek | 28m |
Claims that modern single sign-on traces to an NSA system called Bell Field created during Vietnam, which centralizes keys to issue one-time session keys so… | ||
| How to sell your soul, err, your security program (Jenn Gile) | Jenn Gile | 35m |
Shows how to stop preaching security and start selling outcomes by borrowing product marketing practices: identify the buying committee, run 3–5 interviews per… | ||
| MCPwned: Hacking MCP Servers with One Skeleton Key... (Jonathan Leitschuh) | Jonathan Leitschuh | 45m |
Weaponizes DNS rebinding against SSE/streaming HTTP MCP servers to pivot a browser into local/dev infrastructure, exfiltrates environment variables and… | ||
| Google Drive Hunter: Building the Tool Google Should’ve Built (Ayman Elsawah) | Ayman Elsawah | 23m |
Open-source tool Sleuthther (formerly Google Drive Hunter) scans every Google Workspace user via Directory and Drive APIs, exports per-user sheets with direct… | ||
| Your AI Agent Has Production Access: Now What? (Jack) | Jack | 45m |
Outlines a practical security stack: identify the lethal trifecta (network egress, sensitive-data access, untrusted input) and remove any leg or severely scope… | ||
| nothing matches | ||