← Terug naar nieuwsbrief

AI & Security Nieuws – 28 Januari 2026

Nieuwsbrief

By Djimit* een overzicht voor AI cloud- en security professionals*

De dominante trend vandaag is dat AI zich verder institucionaliseert: nieuwe wettelijke kaders rond generatieve AI worden concreet, terwijl onderzoek focust op efficiëntere, betrouwbaardere modellen en evaluatie­methoden.[1][2][3]

1. Zuid-Korea’s AI Basic Act van kracht

2. EU richting verplichte copyright‑vergoedingen voor genAI

3. Nieuwe evaluatie‑methode voor LLM‑benchmarks (FAQ/PAI)

4. Compacte en robuuste DNN’s via gecombineerde compressie‑ en regularisatie‑technieken

5. Security‑benchmark voor LLaMA‑achtige modellen op OWASP AI Testing Guide

6. AI‑governance: audits bieden schijnveiligheid tegen AI‑aanvallen

7. Nieuwe verantwoordelijkheden voor data‑engineers in AI‑tijdperk

8. Strategische analyses: AI‑regulering en machtsverschuiving in de EU

Insight of the Day

De combinatie van Zuid‑Korea’s verplichte AI‑labeling, Europese plannen voor verplichte copyright‑fees en nieuwe LLM‑evaluatie‑ en security‑benchmarks illustreert dat AI in 2026 verschuift van “bigger models” naar een infrastructuur waarin transparantie, meetbaarheid en afdwingbaarheid de kern‑innovaties worden.[3][4][6][1]

Bronnen[1] Efficient Evaluation of LLM Performance with Statistical … https://arxiv.org/pdf/2601.20251.pdf[2] Towards Compact and Robust DNNs via Compression- … https://arxiv.org/pdf/2601.20301.pdf[3] South Korea’s Landmark Laws Require Labels on AI- … https://petapixel.com/2026/01/29/south-korea-launches-landmark-laws-requiring-labels-on-ai-generated-content/[4] EU advances mandatory copyright fees for AI companies https://www.chosun.com/english/industry-en/2026/01/29/ZCAGCRLN2ZEWRE2XXQCNVCKLGQ/[5] In 2026, AI will move from hype to pragmatism https://techcrunch.com/2026/01/02/in-2026-ai-will-move-from-hype-to-pragmatism/[6] Benchmarking LLAMA Model Security Against OWASP Top … https://arxiv.org/pdf/2601.19970.pdf[7] Passing Audits Won’t Protect You From AI Attacks https://community.ibm.com/community/user/blogs/filip-piletic/2026/01/29/passing-audits-wont-protect-you-from-ai-attacks[8] New Responsibilities For Data Engineers In The Age Of AI https://www.forbes.com/councils/forbestechcouncil/2026/01/29/new-responsibilities-for-data-engineers-in-the-age-of-ai/[9] EU Tech Policy in 2026: From Paperwork to Power Politics https://www.disinformationindex.org/blog/2026-01-29-eu-tech-policy-in-2026-from-paperwork-to-power-politics/[10] 2026 Large Language Models Outlook https://am.vontobel.com/en/insights/2026-large-language-models-outlook[11] AI IN 2026: WHEN THE HYPE MEETS HARD REALITY https://www.faf.ae/home/2026/1/4/the-intelligence-reckoning-artificial-intelligences-brutal-collision-with-reality-in-2026[12] LLM.co Launches Private LLM Infrastructure Purpose-Built for … https://markets.businessinsider.com/news/stocks/llm-co-launches-private-llm-infrastructure-purpose-built-for-cybersecurity-teams-1035758177[13] AI by AI Weekly Top 5: December 29, 2025 – January 4, 2026 https://champaignmagazine.com/2026/01/04/ai-by-ai-weekly-top-5-december-29-2025-january-4-2026/[14] Latest AI News & Technology Updates – January 2026 https://www.vtnetzwelt.com/ai-development/latest-tech-ai-updates-what-january-2026-is-really-telling-us/[15] ByteDance and Alibaba to release new AI models for Lunar … https://ca.finance.yahoo.com/news/bytedance-alibaba-release-ai-models-155211609.html[16] AI News & Trends January 2026: Complete Monthly Digest https://www.humai.blog/ai-news-trends-january-2026-complete-monthly-digest/[17] 2026 Will Change AI Forever | Here’s Why https://www.youtube.com/watch?v=Wf_Sh69MLOQ[18] Advanced Generative AI Course for Developers and … https://www.globenewswire.com/news-release/2026/01/29/3228957/0/en/Advanced-Generative-AI-Course-for-Developers-and-Engineers-Interview-Kickstart-s-New-2026-Program-Focused-on-Model-Development-and-Applied-AI-Systems.html[19] AI in Jan. 2026: Three Critical Global Decisions https://etcjournal.com/2025/12/28/ai-in-jan-2026-three-critical-global-decisions-global-ai-operating-system/[20] 2026 AI Symposium | U.S. Space & Rocket Center https://www.rocketcenter.com/2026-AI-Symposium[21] Ubitus Partners with Maizuru City, Kyoto to Launch AI Data … https://www.afp.com/en/infos/ubitus-partners-maizuru-city-kyoto-launch-ai-data-center-project-advancing-deployment-top[22] Artificial Intelligence Jan 2026 – arXiv https://arxiv.org/list/cs.AI/current[23] January 29, 2026 – Artificial Intelligence Workshop – AFCEA https://afcea-la.org/events/january-29-2026-ai-workshop/[24] Meta Forecasts Spending of at Least $115 Billion This Year https://www.nytimes.com/2026/01/28/technology/meta-earnings-ai-spending.html[25] What Are the Barriers to AI Adoption in Cybersecurity? https://www.paloaltonetworks.ca/cyberpedia/what-are-barriers-to-ai-adoption-in-cybersecurity[26] The Sunday Daily: Hollywood’s A.I. Moment https://www.nytimes.com/2026/01/18/podcasts/the-daily/the-sunday-daily-hollywoods-ai-moment.html[27] I built marshmallow castles in Google’s new AI world … https://techcrunch.com/2026/01/29/i-built-marshmallow-castles-in-googles-new-ai-world-generator-project-genie/[28] Medical chatbot using OpenAI’s GPT-3 told a fake patient to … https://go.nature.com/3p9HRoY[29] Artificial Intelligence (AI) Dataset Search Platform https://www.globenewswire.com/news-release/2026/01/29/3228443/0/en/Artificial-Intelligence-AI-Dataset-Search-Platform-Research-Report-2025-5-66-Bn-Market-Opportunities-Trends-Competitive-Analysis-Strategies-and-Forecasts-2019-2024-2024-2029F-2034F.html[30] Trump and States Aim to Stop A.I. From Inflating Energy Bills https://www.nytimes.com/2026/01/16/business/trump-ai-electricity-costs.html[31] Using AI to combat hate speech http://go.nature.com/4h5fv7w[32] Amazon to Cut 16000 Jobs in Latest Round of Layoffs https://www.nytimes.com/2026/01/28/technology/amazon-corporate-layoffs.html[33] OpenAI Developer Community https://community.openai.com/top[34] Opportunity Costs of State and Local AI Regulation https://www.cato.org/policy-analysis/opportunity-costs-state-local-ai-regulation[35] Large Language Models – AI-Powered arXiv Paper Summarization https://www.summarizepaper.com/en/arxiv-id/2402.06196v1/[36] OpenAI’s first piece of hardware https://community.openai.com/t/openai-s-first-piece-of-hardware/1367439[37] Exploring AI Strategies for Solving NYT Letter Boxed More … https://community.openai.com/t/exploring-ai-strategies-for-solving-nyt-letter-boxed-more-efficiently/1372830[38] Training Reasoning Models on Saturated Problems via … https://arxiv.org/pdf/2601.20829.pdf[39] Tips, Tools, Techniques, and GPT Integration for Gamedev … https://community.openai.com/t/ai-in-game-development-tips-tools-techniques-and-gpt-integration-for-gamedev-success/1372841[40] Protect copyrighted work used by generative AI, say Legal … https://europeansting.com/2026/01/29/protect-copyrighted-work-used-by-generative-ai-say-legal-affairs-meps/[41] Gamedev Tips, Tools, Techniques, and GPT / LLM Agent … https://community.openai.com/t/ai-in-game-development-gamedev-tips-tools-techniques-and-gpt-llm-agent-integration/1372841[42] Artificial Intelligence https://arxiv.org/list/cs.AI/recent

Capability overhang

Capability overhang verwijst naar de kloof tussen wat frontier AI-systemen technisch kunnen en hoe gebruikers, bedrijven of landen deze capaciteiten daadwerkelijk benutten. Bij bedrijven uit zich dit vooral in onderbenutting van geavanceerde functies zoals chain-of-thought prompting, tool-integratie of diepgaande reasoning, ondanks de snelle groei van modelprestaties.[1][2][3][4]

Definitie

OpenAI definieert capability overhang als de “widening distance between what frontier AI can do and the extent to which those capabilities are being used” door typische gebruikers. Het gaat om een mismatch: modellen verdubbelen capaciteiten (bijv. task horizons elke 7 maanden sinds 2019), maar workflows blijven oppervlakkig. Bedrijven vangen slechts een fractie van de mogelijke waarde op, vergelijkbaar met ongebruikte servercapaciteit pre-cloud.[5][6][1]

Bedrijfscontext

Bij bedrijven is de kloof meetbaar via “thinking capabilities”: een proxy voor modelinspanning per query (diepte van gebruik). Power users (vaak tech-savvy teams) benutten 7x meer dan gemiddelde gebruikers; in de EU adopteert slechts 17% van mkb’s AI versus 55% van grote ondernemingen. Oorzaken zijn skilltekorten, ROI-metingproblemen, regelgevingsonzekerheid en data-kwaliteitsissues.[7][1]

AspectPower users / KoplopersGemiddelde / AchterblijversThinking capabilities7x meer per gebruikerBasisgebruik (geen advanced prompting) [1]Adoptie mkb vs groot55% groot17% mkbProductiviteitswinstVolledig benut (coderen, onderzoek) [8]Gekoppeld aan een paar sectoren/firma’s

Impact op Bedrijven

Zonder reductie concentreert productiviteit zich bij weinigen, wat concurrentievoordeel creëert voor early adopters maar risico’s op ongelijkheid. OpenAI raadt training (bijv. SME AI Accelerator voor 20k EU-mkb’s), meetbare adoptie-doelen en skills-accreditatie aan om dit te dichten. Dit verschuift focus van modelgrootte naar gebruiksdiepte.[2][7]

Bronnen[1] OpenAI’s Capability Overhang: A Market Analysis for 2026 https://www.aicerts.ai/news/openais-capability-overhang-a-market-analysis-for-2026/[2] Organizational adoption of AI: capabilities vs. uses https://www.duperrin.com/english/2026/01/26/ai-agentic-uses-capability-overhang/[3] The next chapter for AI in the EU https://openai.com/index/the-next-chapter-for-ai-in-the-eu/[4] OpenAI Launches EU SME AI Accelerator for 20,000 Businesses https://www.resultsense.com/news/2026-01-28-openai-launches-eu-sme-ai-training-programme[5] Ending the Capability Overhang https://cdn.openai.com/pdf/openai-ending-the-capability-overhang.pdf[6] EU Economic Blueprint 2.0 https://cdn.openai.com/pdf/openai-eu-economic-blueprint-jan-2026.pdf[7] OpenAI Launches EU Economic Blueprint 2.0 to Boost AI … https://enterpriseai.economictimes.indiatimes.com/news/industry/openai-launches-eu-economic-blueprint-20-to-boost-ai-adoption-in-europe/127770641[8] OpenAI CFO Sarah Friar: There’s a ‘mismatch’ between AI’s … https://fortune.com/2026/01/27/openai-cfo-sarah-friar-mismatch-ai-abilities-value-companies-capturing/[9] How countries can end the capability overhang https://openai.com/index/how-countries-can-end-the-capability-overhang/[10] Breakingviews – OpenAI’s hype machine faces a corporate challenge https://www.reuters.com/commentary/breakingviews/openais-hype-machine-faces-corporate-challenge-2025-10-28/[11] There’s a massive ‘capability overhang’ in AI that’s occurring https://www.youtube.com/watch?v=tUVSuFT301U[12] What we can learn from AI power users about closing the … https://www.linkedin.com/pulse/what-we-can-learn-from-ai-power-users-closing-chatterji-yxpoc[13] Chris Lehane’s Post https://www.linkedin.com/posts/chris-lehane-2562535_weve-released-a-new-eu-economic-blueprint-activity-7422613923922788352-L0ZQ[14] There’s a massive ‘capability overhang’ in AI that’s occurring https://www.reddit.com/r/aiwars/comments/1qjge2f/openai_cfo_sarah_friar_theres_a_massive/[15] AI for self empowerment https://openai.com/index/ai-for-self-empowerment/

Recente Hoog-Impact Items

Hoewel geen items exact op 28 januari 2026 vallen, zijn dit de meest recente meldingen uit erkende bronnen (27-28 januari).

Actief Geëxploiteerde Zero-Day

Fortinet meldt CVE-2026-24858, een kritiek authenticatie-bypass lek in FortiOS/FortiCloud SSO, actief geëxploiteerd door malafide accounts tot 22 januari; patches beschikbaar, CISA verplicht update voor 30 januari.[3][1]

APT-Campagne Update

Mustang Panda (Chinese APT) deployt geüpdatete COOLCLIENT backdoor voor browser-data diefstal en clipboard monitoring in overheidsaanvallen; geen specifieke CVE, wel MITRE-compatibele TTPs zoals credential access.[2][4]

Groot Datalek

SoundCloud: 29,8 miljoen accounts met persoonlijke/contactgegevens gestolen; geen CVE, maar infostealer-gerelateerd, gemeld 27 januari.[5][2]

Bronnen[1] Fortinet waarschuwt voor actief aangevallen lek in FortiOS en … https://www.security.nl/posting/922411/Fortinet+waarschuwt+voor+actief+aangevallen+lek+in+FortiOS+en+andere+producten[2] All News – Bleeping Computer https://www.bleepingcomputer.com/news/[3] CVE-2026-24858: FortiOS SSO Zero-Day Exploited in the Wild https://socprime.com/blog/cve-2026-24858-vulnerability/[4] Mustang Panda Deploys Updated COOLCLIENT Backdoor in … https://thehackernews.com/2026/01/mustang-panda-deploys-updated.html[5] Bijna 30 miljoen SoundCloud-accounts getroffen door datalek https://www.techzine.nl/nieuws/security/574453/bijna-30-miljoen-soundcloud-accounts-getroffen-door-datalek/[6] CVE-2026-21509: Actively Exploited Microsoft Office Zero-Day Forces Emergency Patch https://socprime.com/blog/latest-threats/cve-2026-21509-vulnerability/[7] Actuele Cyberaanvallen / Menu Nieuws & Trends | Cybercrimeinfo.nl https://www.ccinfo.nl/menu-nieuws-trends/actuele-cyberaanvallen[8] Cisco confirms active exploitation of new Zero-Day in HTTP … – Reddit https://www.reddit.com/r/cybersecurity/comments/1qmgiw7/cisco_confirms_active_exploitation_of_new_zeroday/[9] Global cyber threat campaigns escalate as APT groups target critical … https://industrialcyber.co/ransomware/global-cyber-threat-campaigns-escalate-as-apt-groups-target-critical-sectors-intel-471-reports/[10] Datalek met 149 miljoen inloggegevens van Facebook, Netflix … https://techpulse.be/nieuws/494320/datalek-met-149-miljoen-inloggegevens-van-facebook-netflix-gmail-en-andere/[11] EU Cyber Resilience Act: Key 2026 milestones toward CRA … https://www.jdsupra.com/legalnews/eu-cyber-resilience-act-key-2026-1304993/[12] Security.NL https://www.security.nl[13] Earth Kurma APT is actively targeting government and … https://securityaffairs.com/177125/apt/earth-kurma-apt-is-actively-targeting-government-and-telecommunications-orgs-in-southeast-asia.html[14] Data Protection Day 2026 | Brussels Hoofdstedelijk Gewest https://be.brussels/nl/over-het-gewest/data-protection-day-2026[15] European Commission proposes revised Cybersecurity Act to boost … https://industrialcyber.co/regulation-standards-and-compliance/european-commission-proposes-revised-cybersecurity-act-to-boost-eu-cyber-resilience-secure-ict-supply-chains/[16] Altman van OpenAI en Cook van Apple uiten kritiek op ICE na … https://nl.marketscreener.com/beursnieuws/openai-s-altman-zegt-tegen-medewerkers-ice-gaat-te-ver-na-dodelijke-incidenten-in-minnesota-ce7e5bd9d98cf525[17] Microsoft Patch Tuesday addresses 112 defects, including one … https://cyberscoop.com/microsoft-patch-tuesday-january-2026/[18] Microsoft January 2026 Patch Tuesday fixes 3 zero-days, 114 flaws https://www.bleepingcomputer.com/news/microsoft/microsoft-january-2026-patch-tuesday-fixes-3-zero-days-114-flaws/[19] January 2025: Recent Cyber Attacks, Data Breaches, Ransomware … https://www.cm-alliance.com/cybersecurity-blog/january-2025-recent-cyber-attacks-data-breaches-ransomware-attacks[20] Microsoft’s January 2026 Security Blitz: 114 Patches, Zero Days, and … https://www.spreaker.com/episode/microsoft-s-january-2026-security-blitz-114-patches-zero-days-and-ai-marketplace-explosion–69486662[21] The Most Recent Data Breaches in 2026 – Breachsense https://www.breachsense.com/breaches/[22] The State of Incident Response 2026: Insights from 630 Cyber … https://www.eye.security/blog/the-state-of-incident-response-2026-insights-from-630-investigations[23] January 2026 Microsoft Patch Tuesday: Actively exploited zero day … https://www.csoonline.com/article/4116437/january-2026-microsoft-patch-tuesday-actively-exploited-zero-day-needs-attention.html[24] Data Protection Day 2026: Reset or refine? https://www.edps.europa.eu/data-protection/our-work/publications/events/2026-01-28-data-protection-day_en[25] Als gedupeerde zie ik niets terug van de aan een online dienst … https://www.security.nl/posting/922466/Als+gedupeerde+zie+ik+niets+terug+van+de+aan+een+online+dienst+opgelegde+boete+wegens+een+datalek_+Wat+kan+ik+daar+aan+doen%3F[26] Microsoft investigates Windows 11 boot failures after January updates https://www.bleepingcomputer.com/news/microsoft/microsoft-investigates-windows-11-boot-failures-after-january-updates/[27] Allianz Risk Barometer 2026 – Cyber incidents https://commercial.allianz.com/news-and-insights/expert-risk-articles/allianz-risk-barometer-2026-cyber-incidents.html[28] Zero-Day Exploits Surge, 30% of Flaws Attacked Before Disclosure https://www.infosecurity-magazine.com/news/zeroday-exploits-surge-vulncheck/[29] Data Protection Day – 28 January 2026 – Registration form https://www.edps.europa.eu/webform/data-protection-day-2026_en

DjimIT Nieuwsbrief

AI updates, praktijkcases en tool reviews — tweewekelijks, direct in uw inbox.