schedule_2026

We all know Quora - or do we? Like Internet herpes, if you're a Google user, it'll follow you, forever, haunting your inbox with clickbait if you Google while logged into your account. ...But do we REALLY know Quora? Because despite being a pustule on the Internet that exists for the sole purpose of spamming SERPs, you'll be shocked to learn that it's also got - spoiler alert!!! - a dark, seedy underbelly. One that I uncovered while volunteering on an MP investigation. One that can yield surprising, disturbing and useful intelligence. In this talk, I'll explain: 1. The traces Quora leaves behind when something is "limited," "deleted" or a user is "banned." Because on Quora, 'deleted' just nulls the post body while the API keeps serving the slug and author, and 'limited' barely hides anything at all. On Quora, Limited is UNLIMITED, just like Olive Garden's breadsticks! Except unlike Olive Garden breadsticks, Quora's "limited" option is a fig leaf, and "deleted" content isn't much better: The API doesn't hide it. It coughs up the slug and author fully visible to other accounts and even logged-out strangers. 2. How to use Quora's API hairball to see what a user posted and build a network graph 3. What the disturbing subcultures on Quora mean for OSINT 4. Limitations of approach and ideas for automating OSINT Trigger warnings: This talk may include mention of disturbing topics, though all information will be anonymized / sanitized and no graphic content shared.

Nico Dekens

The OSINT landscape is undergoing a fundamental shift. Artificial Intelligence is no longer just a tool; it is becoming a crutch. As analysts increasingly rely on LLMs to triage data, translate foreign slang, and summarize massive datasets, a dangerous cognitive atrophy is taking hold. We are trading the slow, deliberate friction of critical thinking for the illusion of speed and certainty. This talk, "OSINT Is Still a Thinking Game," exposes the hidden vulnerabilities of AI dependence in intelligence work. Through real-world case studies—from misinterpreting Telegram chatter to falsely flagging logistics data—we will examine how the "Good Enough" trap leads to catastrophic analytical failures. More importantly, this session provides a concrete defense framework. Attendees will learn the four essential habits required to survive the AI era: reading the raw source, forcing a second hypothesis, separating speed from confidence, and auditing dependence. The tools will inevitably get better, but the thinking must get sharper. Join this session to learn how to maintain your tradecraft, keep your judgment visible, and ensure that in the age of automation, defensibility always wins over speed.

Anthony Russell

Most LLM-powered recon tools look impressive in a 90-second demo and fall apart on real targets. They hallucinate, loop, go out of scope, double-fire the same endpoint, or die the moment a WAF or rate limit appears. This talk is about what it actually takes to run fifty-plus specialist agents in parallel for hours or days without the chaos. We built a system where every agent is locked to the rails: a concrete tool, a strict pipeline phase, explicit prerequisites, timeouts, and sandboxes. The model does not drive. It rides. The binaries (and only the binaries) touch the target. This "Agents on Rails" approach eliminates freestyle LLM behavior while still letting the model do what it is good at—reasoning over evidence.

Eddie Miro

We live in an era where our location data is constantly harvested, but what happens when the tracking becomes physical? From the clandestine beacons of the Cold War to the consumer-grade AirTags tucked into backpacks today, physical tracking technology has become incredibly cheap, accessible, and pervasive. In this talk, we will trace the evolution of physical tracking tech, analyze how modern implementations exploit wireless protocols like BLE, cellular, and GPS, and discuss practical defense strategies to detect and neutralize unwanted eyes. Finally, we will demystify the threat by turning the tables: demonstrating how to build and deploy a fully functional, budget-friendly tracking beacon using off-the-shelf DIY hardware. Attendees will leave with a deep understanding of the tracking landscape and the knowledge required to both defend against and build these systems.

Most threat intelligence focuses on what attackers have already done: payloads, malware families, and post-compromise behavior. But adversaries are far more fluid at the payload layer than they are at the infrastructure layer. Domains rotate, IPs churn, and malware gets recompiled but infrastructure leaves patterns. This article explores how to track threat actors through their infrastructure by leveraging fingerprinting techniques that expose those patterns at scale. We’ll walk through practical methods including JARM for active TLS stack fingerprinting, JA3/JA4 for identifying consistent communication behaviors, SSH host key correlation for infrastructure pivoting, and MurmurHash for clustering phishing kits and web panels through shared assets. Individually, these signals are useful. Combined, they allow defenders to map infrastructure clusters tied to a single actor or campaign—even when traditional indicators change. The focus is not on attribution for its own sake, but on building a repeatable approach to discovering “sister infrastructure” and identifying campaigns earlier in their lifecycle. By shifting attention away from payloads and toward the systems attackers stand up to operate, defenders can detect patterns before deployment and reduce time to awareness. Attackers rely on reuse of configurations, tooling, and infrastructure. That reuse creates fingerprints. And those fingerprints are often more durable than the indicators most teams prioritize. If you want to track threat actors effectively, stop chasing malware. Track the infrastructure they can’t help but reuse.

Sadettin Boluk

Modern Wi-Fi clients are designed to hide. They randomize MAC addresses, reduce directed probes, avoid exposing preferred networks, and use per-network private addresses. Yet before a device connects, it still speaks. This talk introduces a Python-based proof-of-value tool for passive Wi-Fi reconnaissance and privacy exposure analysis. The tool listens to 802.11 management traffic, builds an environmental AP map from beacon frames, observes client reactions through probe requests, parses Information Elements from probe and association frames, and correlates randomized MAC identities using IE semantics, sequence behavior, packet size, timing, and channel context. The core idea is, even when the MAC address changes, the device’s wireless behavior may remain linkable. We will demonstrate how passive wireless metadata can reveal device presence, movement context, SSID exposure, privacy leakage, and probable same-device candidates even when MAC randomization is enabled. The demo will use only controlled test devices in a lab environment. The talk also introduces an AI-assisted scoring module trained on IE-level Wi-Fi fingerprints to improve correlation accuracy and reduce false positives. By combining semantic 802.11 features with behavioral signals, the tool aims to produce a practical privacy exposure score for Wi-Fi clients. This is not a tracking product or a vendor-specific platform. The goal is to show defenders, researchers, privacy engineers, and red teams what Wi-Fi clients still expose by default, and how passive management-frame metadata can become a meaningful reconnaissance surface.

Eli Woodward

Cl0p has executed at least nine major exploitation campaigns since 2020, targeting file transfer and edge devices from Accellion to MOVEit to Centrestack. Each campaign looks different on the surface, different CVEs, different victims, different tooling. But their infrastructure tells a different story. This talk presents the results of a multi-year infrastructure reconnaissance effort tracking Cl0p's hosting, ASN usage, and operational patterns across all known campaigns. By mining passive DNS data, network telemetry, and hosting records, I mapped the infrastructure behind each campaign and identified patterns the group can't seem to shake — including a favorite bulletproof hosting provider used across four separate campaigns, a finding that roughly one-third of their ASNs get recycled, and pre-attack reconnaissance probing observed as far as two years before exploitation.

What if your recon tool could understand "find leaked credentials for this company on the dark web, cross-reference with their exposed infrastructure, and show me the attack path" — and then actually do it, autonomously, with zero manual commands? CyberKB is a 214,000-line open platform that puts an AI copilot (Keke) in command of 131 offensive tools spanning reconnaissance, dark web OSINT, Shodan intelligence, breach databases, and a full Kali Linux shell — all orchestrated through natural language conversation. The Dark Web Intelligence Pipeline (DarkMonitor): CyberKB's DarkMonitor module queries .onion search engines concurrently through Tor, uses LLM-powered query refinement to generate optimal dark web search terms, applies AI relevance filtering to surface the most critical results. The AI Copilot (Keke): Keke isn't a wrapper around ChatGPT. It's a function-calling agent with direct execution access to: a privileged Kali container (nmap, curl, nikto, sqlmap, gobuster, hydra — the full toolkit), 16 dark web search engines via Tor, breach and paste database aggregators, a RAG-powered knowledge base with semantic search over ingested recon data, Shodan lookups and CVE correlation, MITRE ATT&CK technique mapping, attack graph generation and path prioritization, and engagement management (targets, credentials, findings, reports). Scale Proof — 2,250-Domain Assessment: We demonstrate CyberKB against a real-world external attack surface assessment: 2,250 domains belonging to a single organization, producing 11,968 unique IPs, 8,494 hostnames, 2,444 CVEs, 80,238 open ports, and a knowledge graph of 6,390 nodes with 51,990 infrastructure relationship edges. The entire dataset was parsed, correlated, and ingested into the AI-queryable knowledge base in 25 seconds. Keke can now answer questions like "which x domains share infrastructure with known-vulnerable IPs" by searching the graph, not by re-scanning. What This Means for Defenders: Every autonomous recon step Keke performs leaves detectable artifacts. We discuss what blue teams should monitor: DNS burst patterns from multi-engine enumeration, Tor exit node correlation with target infrastructure, behavioral signatures of AI-driven sequential probing, and how to distinguish human recon from autonomous agent recon. The same platform that empowers red teams reveals the detection surface that defenders can leverage. Key Takeaways: 1. AI copilots with direct tool execution compress hours of reconnaissance into minutes of conversation — fundamentally changing the operator's role from command executor to strategic decision-maker. 2. Dark web OSINT can be systematically automated with LLM-driven search refinement and relevance filtering, making it accessible beyond specialized analysts. 3. Infrastructure-scale attack surface mapping (2,250+ domains) becomes practical when AI handles correlation instead of humans. 4. Autonomous recon creates detectable patterns that blue teams should be actively hunting for. Tool Stack: Python, Flask, ChromaDB (RAG), Docker (Kali + Tor + Browser Agent), OpenAI-compatible LLM API, SQLite, Playwright, BeautifulSoup. 105 Python modules. Fully self-hosted — no cloud dependencies for core functionality.

Avishai Efrat

Roey Ben Chaim

AI agents quietly created a new external attack surface: copilots, custom agents, AI cakends and various deployments that ship to the internet, often without anyone realizing they are reachable, enumerable, or over-permissive. In this talk, we’ll show how attackers can already find your agents in the wild, shedding light on the technical details that enable this kind of malicious activity, including how we used these details to find 1000s of exposed agents of different kinds. We’ll follow up with explaining how to measure exposure, see the proof for obscurity failing, and understand how to detect threat-actor agent-focused recon before it turns into an impactful attack. Capping it all off by showcasing PowerPwn, a recon tool you can use to test your own exposure

John McCary

We scanned every IPv4 address on the planet. Twice. We collected public records in one of at least three places: the RIR registration, the operator's reverse DNS, or whatever service answers on port. Most threat intelligence pipelines read one of those. Darkmouse lines up all of them and flags the disagreements. The result is attack-surface visibility that no single source gives you, and a methodology you can run yourself. This talk walks the recon pipeline behind three findings from a single cross-domain pass over 4.3 billion IPv4 addresses in five days, plus a 92 million IP targeted scan in 34 hours (Russia, Iran, and North Korea). We used ZMap on single-use Jetstream2 VMs, zdns for PTR and forward verification, bulk RPSL and ARIN XML and APNIC data loaded into DuckDB alongside MaxMind GeoLite2, OpenSanctions, and the US Trade Consolidated Screening List. Enrichment is baked into the scan row at ingest. Every field carries a source-date stamp so longitudinal comparison actually works. Lead finding for a recon audience: 6,656 Iranian IPs in RIPE where the registrant placed fabricated US street addresses into the authoritative registry. 4,608 of them cite AT&T Mobility's ARIN IP-management address verbatim. 2,048 cite a Philadelphia apartment building. All geolocate to Iran. All trace to one operator through a shared RIPE maintainer object linking a Shiraz Local Internet Registry and an Omani shell. Between September 2025 and January 2026 the fabricated addresses began rotating out, replaced by netnames like "Datacamp-Limited" that impersonate real UK hosting companies. A single-point-in-time feed cannot see that rotation. Two dated snapshots can. We are currently running follow up scans and expect to have new data before the conference. Additional Findings: five active PTR records in Russia and the Netherlands claiming US government domains, including .fbi.gov subdomains, four of five still live six months after first observation (April 2026). And 1,534 APNIC-registered IPs for Entity Listed Chinese telecoms, declared country=US, geolocating to One Wilshire in Los Angeles, running production email and DNS on FCC-revoked Section 214 infrastructure. Every finding ships with the RDAP query, the DNS check, and the cross-reference that verifies it. Every finding ships with the RDAP query, the DNS check, and the cross-reference that verifies it. Attendees leave knowing which four sources to line up, which fields to trust, and what to look for when two dated snapshots disagree.

Michael Reimsbach

Rishi C

People are getting hired and trusted every day. Some of them do not exist at all, yet they still pass interviews, collect paychecks, and gain access to sensitive systems. Campaigns attributed to the DPRK have shown that this threat is very real. So how do you catch a ghost with a resume? Attendees will learn practical OSINT techniques for spotting fake personas and receive a checklist for thorough background checks. They will see these methods applied through two cases based on a true story, illustrating how these personas succeeded, how one could have been prevented, and where OSINT reaches its limits. These techniques not only help attendees detect fake personas but also provide practical ways to protect their own privacy and control what personal information is visible online.

Jason Haddix

Building AI-Powered Penetration Testing Bots In this talk, we'll walk through the core design philosophy behind AI hackbots and the architecture that makes them work: single-purpose vs. multi-stage bots, context engineering for targeted prompting, and tool integration with output parsing workflows. Then we'll get hands-on. We'll live-build a functional hackbot for a common offensive task — demonstrating asset discovery, endpoint analysis, or mutation-focused testing (e.g., XSS/SSRF) — and show how context engineering and hallucination mitigation work in practice against real targets. You'll see where AI genuinely accelerates reconnaissance and web analysis, and where it falls flat if you aren't careful. We'll close with lessons on cost optimization, auditability, and integrating hackbots into actual engagements. Who should attend: Pentesters and bug bounty hunters with offensive security experience who want to meaningfully integrate AI into their workflow — not as a novelty, but as reliable tooling. What you'll walk away with: A clear mental model for when and how to build hackbots, a live demo you can replicate, and a path into the full course where you'll build seven production-ready bots from asset discovery through mutation testing.

You scrape a public site. You enumerate subdomains. You grep GitHub for secrets. You curl a misconfigured API. Each one touches a statute. Some have been to the Supreme Court. Most operators don't know which is which — until the preservation letter arrives. A federal-court-qualified expert witness (U.S. v. Sullivan) walks through five live OSINT techniques with real-time legal annotation. Zero lawyer-speak. GitHub release included.

Redon Gashi

Armend Gashi

LLM-powered tooling is becoming a force multiplier for attackers, from reconnaissance automation to post-exploitation enumeration. Using their own API keys creates attribution and cost, so they look for alternatives. Thousands of inference endpoints sit exposed on the internet: misconfigured Ollama instances, open vLLM deployments, leaked API keys in directory listings, and expired OAuth tokens from AI coding assistants that can be silently refreshed. Attackers can discover these resources and use them as free compute for their operations, without spending a dollar or registering an account. Attendees will learn how to: •⁠ ⁠Discover exposed inference endpoints and leaked API keys using Infreerence •⁠ ⁠Validate that discovered resources provide usable inference access •⁠ ⁠Operate Echidna, powered entirely by discovered inference •⁠ ⁠Chain LLM skill agents together to execute a guided campaign against a target network Current LLMs are effective force multipliers when directed, and significantly more so when the compute bill goes to someone else. This is resource hijacking applied to the AI era: living off someone else's inference. Understanding this threat is essential for any organization deploying or exposing AI infrastructure. Two tools will be released as open source during the session: •⁠ ⁠Infreerence: A multi-phase scanner and dashboard that discovers exposed inference endpoints and leaked API keys through Shodan and Censys, validates them against live provider APIs, and catalogs usable inference resources across 10+ providers. •⁠ ⁠Echidna: A Mythic C2 agent type that consumes discovered inference endpoints and turns them into operator-directed skill agents for reconnaissance, exploitation planning, post-exploitation, and lateral movement.