Security Café

About Security Café

“Pull up a chair at the Security Café – your monthly deep dive into the hottest cybersecurity news and trends. Each episode brings you an inspiring guest and a story that will spark your imagination. Produced by Quint & Menno (Atos), this is where insights meet conversation. Don’t just stay informed—join the discussion!”

Publisher

Quint Ketting Menno van der Horst

Category

news · technology

Language

en

Explicit

No

First episode

2 Feb 2024

Latest episode

6 Oct 2026

Latest episodes

36 episodes in the feed.

1. 

   6 Oct 2026

   # Three sectors, one crisis — Martin de Vries **Live from CybersecNL — Episode 7 of 7, the finale**

   Note: the video focus slipped during this live recording at the booth. The conversation is the point, and the audio version is unaffected. Recorded at the Atos booth, CybersecNL 2026, minutes after Martin's main stage talk. Martin de Vries has worked in security across three sectors: banking, higher education and industry. He was CISO at a Dutch university when it was hit by a cyberattack, and now works in the industrial sector. Hosted by Menno van der Horst with Quint Ketting. ## Timestamps 00:00 Straight from the main stage 00:50 Three sectors: banking, higher education, industry 01:41 What changes when you move sectors: culture 02:26 What doesn't: listening, relationships, threats and risks 02:50 From highly regulated to compliance as an afterthought 03:20 The CISO as facilitator 04:20 "I own the process, not the risk" 05:28 Why "I told you so" isn't the answer 05:49 Accepting risk at the right level 06:24 Acquisitions, due diligence and security 07:28 The shelved report, and the ransomware that followed 09:42 You don't spend a million to avoid 1.2 10:12 The cyberattack on the university 12:18 A Signal message on a Saturday evening 13:19 Deciding near midnight without the business present 13:43 Having the mandate versus taking it 15:00 IT back online within a week 15:41 Postponing an entire exam week 17:13 Research grant deadlines and month-end salaries 18:20 Disconnecting the network, and session tokens expiring 19:01 Restoring, and in which order to reconnect 19:39 Taking care of your teams 21:59 Sending a whole team home for twelve hours 22:21 Quint: Slammer, and sleeping at the office at twenty 23:08 Room for the personal during a crisis 25:07 Sunday 9 a.m.: what do we share? 25:44 At a university, internal communication is public 26:17 One update per day, at a fixed time 28:32 Openness as part of the culture 29:44 Closing: talk to your counterparts in other sectors ## Key takeaways - Moving sectors changes the culture and maturity you work in. Threats, risks, measures and the need to listen stay the same. - The CISO owns the risk process, not the risk. The job is to find the risk owner and make sure the risk is understood and accepted at the right level. - After an incident, fix first. Then revisit the trade-off between impact and the cost of prevention. - Write down the mandate to disconnect, and make sure the rest of the organisation knows it exists. - Technical recovery within a week came from years of investing in processes. The business impact, from exams to grants to payroll, ran much longer. - Look after your people. Rotate shifts, and when needed send a whole team home. - Communicate once a day at a fixed time, identically internally and externally. It prevents everyone from refreshing for news. - Share what happened, so others don't make the same mistakes. ## Mentioned Fox-IT, which presented the incident analysis Maastricht University, which shared openly after its 2019 ransomware attack ## The full series Ramsés Gallego, Kelvin Rorive, Marianne Schinkel, Aernout Reijmer, Mandy Andress, Gilad Friedman and Martin de Vries. --- SecurityCafe. Powered by Atos.

2. 

   1 Oct 2026

   # Securing agentic AI — Gilad Friedman **Live from CybersecNL — Episode 6 of 7**

   Recorded at the Atos booth, CybersecNL 2026. Gilad Friedman is part of the agentic AI protection team at Radware, a cybersecurity company with nearly 30 years in the business. Hosted by Menno van der Horst with Quint Ketting. Atos and Radware are partners. ## Timestamps 00:00 Recording number seven or eight — nobody is counting anymore 00:35 Gilad: Radware and agentic AI protection 01:40 The first AI risk isn't security. It's governance. 02:05 Then compliance and data leakage, then security 02:40 What people put into AI, and why they bypass the approved tools 03:40 AI-first boards and private accounts at work 04:55 Where are the agents, how many, and what access do they have? 05:15 Shadow AI 05:40 "Our AI policy is simple: it's not allowed" 06:10 Why blocking doesn't work 06:46 A policy is not protection 07:15 One enforcement point instead of twenty vendors 07:40 How do you get visibility on shadow AI? 08:15 Desktop, cloud, mobile and gateway 09:07 The hotel guest, the telecom and legal liability 10:00 On the verge of an explosion 10:30 Connecting AI the way we connected to the internet in the 90s 11:40 Customers: stop the fear, show us enablement 12:45 Build processes that expect the unexpected 13:04 Are today's controls enough for what is already out there? 13:50 What changes is the speed of response 14:20 Prompt injection: planted yesterday, triggered today 15:10 Stopping it, not only detecting it 16:10 Looking back: is your baseline good enough? 16:30 Indirect prompt injection, hallucinations and tool misuse 17:15 Agents inherit the credentials of the user 18:00 How many agents do you have? A silent audience 18:40 Radware drinks its own champagne 19:45 Most of it isn't malicious 20:05 Is credential inheritance simply bad design? 20:30 LLMs can't tell instructions from data 20:55 Infrastructure, architecture, governance, visibility, control 21:25 The major AI incident will come. We just don't know when. ## Key takeaways - The first AI risk organisations run into is governance: they do not know which AI tools and agents exist. Compliance and data leakage come second, security third. - Blocking AI does not stop its use. People move to private accounts, other devices and the web. A policy only works when you can enforce it. - Visibility has to cover desktop, cloud, mobile and the network edge, ideally from one enforcement point rather than a tool per vendor. - Agents inherit the credentials and permissions of the user who runs them. That turns every shared document into a possible attack path. - LLMs cannot distinguish instructions from data. An instruction hidden in an email, file, image or voice message gets executed with the victim's access. - Detection alone is not enough. When an agent asked to review a document starts touching financial systems, it has to be stopped automatically. - Most shadow AI is not malicious. It is people trying to keep up with an AI-first mandate. Enablement works better than fear. ## Mentioned Radware — https://www.radware.com (https://www.radware.com) ## Also in this series Ramsés Gallego, Kelvin Rorive, Marianne Schinkel, Aernout Reijmer and Mandy Andress — out now. Martin de Vries on securing different sectors — the series finale. --- SecurityCafe. Powered by Atos.

3. 

   29 Sept 2026

   # Open source culture — Mandy Andress **Live from CybersecNL — Episode 5 of 7**

   Recorded at the Atos booth, CybersecNL 2026. Mandy Andress is CISO at Elastic, where she has worked for eight years. She was at CybersecNL to deliver a keynote on moving security from human speed to machine speed. Hosted by Menno van der Horst with Quint Ketting. Atos and Elastic work together in several areas, including search and SOC operations. ## Timestamps 00:00 Welcome, third recording of day two 00:16 Mandy: CISO at Elastic, and why she came to CybersecNL 00:58 How Atos and Elastic already work together 01:30 Speed, trust and instinct: incidents at human speed 02:06 Processes built on human delay no longer hold 02:47 The AI-native mindset security needs 03:22 Real-time analysis, and grounding AI with context 04:21 Drinking your own champagne 05:08 Open code, public detection rules, no black box 05:50 Sharing how the security team works, not only what it ships 06:16 "If I keep it to myself, I'm safer" 06:49 Security by obscurity is dead 07:29 Isn't all that openness a risk? 08:15 Knowing your coverage and your gaps as a practitioner 09:05 Security first versus commercial reality 10:14 Eight years at Elastic, and running security on the ELK stack 11:44 "Everything is a search problem" 13:28 Deciding without information: the TU/e breach 13:56 Log4j as a search problem 14:59 The client that went looking with paper and pen 15:40 Never waste a good crisis, and a business that didn't feel the risk 16:10 What if Slammer had targeted Log4j? 16:56 Mandy's keynote: from human speed to machine speed 17:49 Security hasn't changed. The speed has. 18:40 How many agents do you run, and what access do they have? ## Key takeaways - Most security processes are built on the delays that come with human speed. Those delays are gone, and many processes need to be turned around rather than tuned. - Security by obscurity relied on attackers lacking time, imagination or capability. AI agents have all three. - Openness is a security strength. Public code and public detection rules bring in the insight of the global community. - For practitioners, transparency means knowing exactly what a product covers and where you need to augment. A black box leaves you guessing whether you are safe. - Share how you work, not only what you build. Elastic's security team publishes what works, what doesn't, and what it is still figuring out. - Security is a search problem: finding patterns in data. When Log4j hit, knowing where it ran made the difference between acting in minutes and searching by hand. - Security itself hasn't changed. The speed at which we have to act on it has. ## Mentioned Elastic — https://www.elastic.co (https://www.elastic.co) Elastic detection rules — https://github.com/elastic/detection-rules (https://github.com/elastic/detection-rules) Elastic Security Labs — https://www.elastic.co/security-labs (https://www.elastic.co/security-labs) ## Also in this series Ramsés Gallego, Kelvin Rorive, Marianne Schinkel and Aernout Reijmer — out now. Gilad Friedman on AI as an attack vector — coming next. Martin de Vries on securing different sectors — the series finale. --- SecurityCafe. Powered by Atos.

Reach and audience

Public platform figures. Ratings count people who left a rating, not total listeners.

Spotify

5.0 / 5

1 ratings

Contact Security Café

Guest appearances

Books guests

Based on episode analysis; this does not confirm that the show is currently accepting guests.