N
One month in the life of an open-source security project. An AI-driven audit, sent to me privately, reports eleven verified vulnerabilities, including cross-tenant data access and forgeable login tokens.
N
The internet never forgets. We've all heard that before. But what does that really mean? In this talk, we show the chaos, the biggest insights and the weirdest details that come back when you invoke your European right to your own digital history.
N
The published adversarial machine learning literature is mostly wrong about what production AI looks like. The papers describe attacks against undefended classification models in laboratory conditions.
N
GitHub is the most commonly adopted source code management and developer platform in Nordic organizations. However, its attack surface is often neglected or misunderstood. The sheer complexity of all the possible ways you can configure an organization introduces what we’ve identified as almost 50 pitfalls that could lead to compromised source code integrity, or to exposed secrets, tokens and whatever environment it deploys to.
N
AI agents are moving into production fast. They call APIs, read databases, send mail and trigger workflows. In most setups they do this with a service account, a long-lived API key or a token copied from the developer who built them.
N
CSP, Trusted Types, and Fetch Metadata are not checklist headers - they are platform controls that eliminate whole bug classes when you treat them as Secure-by-Design defaults. In this one-hour hands-on workshop we practice the Proactive Security Engineering loop on the browser surface: exploit a class, prove the control, re-test, and leave with a paved-road sketch so the unsafe path is the awkward one - whether the markup was typed by a human or generated by an agent.
N
Ransomware response plans assume the attackers run a clean operation. Across thousands of ransomware cases worked from the front line, roughly one in ten went off track because the threat actors themselves made mistakes.
N
Security vendors now promise AI analysts, autonomous detection, intelligent triage, predictive defense, and machine-speed response. But beneath the label may be a rule engine, a legacy model, an LLM summarizer, or a workflow that still depends heavily on human judgment.
N
In this session, we'll jump into my home lab. No, it's not massive and no, I didn't pay 10.000 Euros to build it, but it works quite well for many tasks.
N
A hands-on session where participants put AI tools under pressure and watch how human behaviour shapes what comes back.
N
Infostealer malware doesn't need your password if it can steal your authentication cookie from disk or memory, and replay it on another machine. Any webserver happily accepts the stolen cookie since it was issued by the webserver itself.
N
Security review at scale has always been a fan-out problem: static analysis, logic review, infrastructure config, policy compliance. In most organizations every one of those fires after you push, when the developer has mentally closed the code days ago.
N
Our AI-assisted security research at AISLE has led to more than 500 CVEs in the last year. LLMs are useful at discovering potential vulnerabilities, but they produce a high volume of candidates that still need investigation.
N
When building a REST API, are you, or the QA specialists in your team, verifying that access policies are satisfied? For every single endpoint? Is it done every single time a new release of the API is made?
N
A hands-on session where participants put AI tools under pressure and watch how human behaviour shapes what comes back.
N
We all love real-time data — clicks, payments, rides, messages — but most of it comes with a catch: it contains personal information we’re not supposed to leak, such as names, emails, locations, or even small clues that can identify someone.
N
An AI code reviewer has to read material supplied through a pull request: code changes, descriptions, images and discussion threads. Some of that material may contain instructions directed at the model.
N
We built an AI-agent pipeline that audits the open-source Go libraries our two products depend on, from picking targets to reporting upstream. Agents shortlist dependencies that parse untrusted input but are not fuzzed upstream (OpenSSF Scorecard helps), write harnesses with proven positive controls, and catch far more than crashes: bounded-work limits for a DoS that never crashes, round-trip and differential parser checks, fail-closed invariants for silently flipped verdicts.
N
A production database deleted by Claude in nine seconds flat, all because of a stale API token in a forgotten file. A typical email phishing attempt turned data exfiltration disaster, all because the malicious actor got their hands on GitHub Copilot.
N
A vulnerability can be public and a patch available while the people who need it remain unaware. Through Shai-Hulud, TeamPCP and PHP2Shell, this talk explores how security knowledge travels through social networks, earns trust and becomes action, and how AI changes the race between attackers and defenders.
N
A Formula 1 team can replace four tires, inspect the car, and send a driver back onto the track in under three seconds. Meanwhile, when I bring my own car to the garage for four new tires, it takes nearly an hour.
N
Most advice on agent security assumes one model with a few tools. A lot of teams are now building systems where agents pass work to each other. A planner hands a task to a research agent, that agent calls a tool, and the result ends up with another agent that can actually change something.
N
Over the past year, I discovered and responsibly disclosed multiple critical cross-tenant vulnerabilities in the same Azure service area, resulting in $200,000 in bounty awards. This work was done as a side project and demonstrates that impactful cloud findings can come from a disciplined methodology for anyone willing to invest time in understanding these systems.
N
Every offensive security engineer has the physical engagement daydream. The badge clone, the tailgate, the server room at 2am. Mine showed up as a solo assignment with one week of notice, two client campuses fifty miles apart, five or more buildings on each, and about a day and a half of usable recon time.
N
A wifi network name that roots your router. A TLS certificate field that takes over hosting accounts. A DNS response that lets you disrupt the routing of a European network. Not your every day injection vectors, and they rarely get the scrutiny they deserve, even in essential infrastructure.
N
Every day, cybersecurity professionals make trust decisions.
N
This talk presents a security evaluation of the QNX Neutrino 7.1 kernel attack surface from user space. QNX is an RTOS deployed in over 255 million vehicles. This research was conducted as part of a pre-integration automotive equipment assessment.