TIKOS®

Pleased to share that TIKOS® has been selected for Cohort 10 of The University of Edinburgh AI Accelerator. A Great couple of days in Edinburgh meeting the Bayes Centre team, programme partners and the rest of the cohort in person. Over the coming months we'll be working on growth strategy and raising investment as we bring our AI Assurance solutions to market. Thanks to Fabrizio Formichella and Sarah Hall for putting together a great start to the programme. #AIAccelerator #Edinburgh #Founders

TIKOS® has been selected for Wavestone's 'UK Cybersecurity Startup Radar 2026', in the Securing AI category, recognising UK innovators building the security layer for AI systems. It's a good moment to share a first look at what's next. TIKOS® CyberProfiler, our first standalone AI cyber security product, launches officially at AMLUCS on 23 September, the UK's leading AI/ML/cyber security conference (Dstl-sponsored); where we also have a presentation slot and poster. TIKOS® is the only solution that can access and analyse the internal decisions of deep-learning models (including LLMs) at run-time. Competitors are limited to observing model inputs and outputs. TIKOS® CyberProfiler detects prompt injection (the underlying risk behind many of OWASP GenAI Security Project's LLM and Agentic Top 10's) by profiling the intent behind a prompt, not scanning it for banned words. Because it works from a signature of intent rather than vocabulary, it catches attack types it has never seen before, without waiting for a rule or patch update. The technique and technology has been validated in a Proof of Technology project with Cisco and the National Cyber Security Centre (NCSC is part of GCHQ), with 100% success rate, zero false positives and an inference classification latency of under 35ms, ideal for an inline guardrail used in high stakes and safety critical environments. More to come as we approach the AMLUCS launch. Build Trustworthy AI. In great company in this segment with: Quantra, CybPass, SPLX, a Zscaler Company, VerifyWise, AquilaX Security, ETrust Labs, Secure Agentics, CloudPeek AI, Open Code Mission, norse3, Offset Labs, Cyber AI Trust, DevSecAI, CattleGrid, Zerodai, LLM Pirate - and many more. https://lnkd.in/efNzhMTD #AIsecurity #Cybersecurity #PromptInjection #TrustworthyAI #AMLUCS

We're presenting at AMLUCS 2026! (Applied Machine Learning for Cyber Security Conference) 23-24 September 2026, Queen Elizabeth Olympic Park, London, E15 2GW AMLUCS brings together practitioners, researchers, policymakers and operators working at the intersection of AI and cybersecurity. TIKOS® is proud to be included in this year's Talks, Keynotes and Technical Sessions track, alongside BAE Systems, Dstl, Microsoft, Palo Alto Networks, The Alan Turing Institute and many others. Our session, "Reasoning, Not Rationalising: Mechanistic Guardrails for Agentic AI in Critical National Infrastructure", shares results from a proof-of-technology project delivered with Cisco and the National Cyber Security Centre (NCSC is part of GCHQ), and which achieved perfect control, zero false positives, zero false negatives, at under 35ms end to end latency for each agent input. Prompt injection is the dominant unsolved threat to agentic AI. Blocklists and output filters are reactive, new phrasing means a new bypass. Because TIKOS® accesses a model's internal decisions, not just its inputs and outputs, we profile the causal reasoning pathway itself; watching for the mathematical signature of a guardrail bypass, rather than scanning text for forbidden words. Full programme details to follow from the AMLUCS organisers https://www.amlucs.ai/ Build Trustworthy AI. #AMLUCS #CyberSecurity #AgenticAI #CriticalInfrastructure #AISecurity

We're proud to be one of companies selected for FinTech Scotland's Financial Regulation Innovation Lab (FRIL): Responsible AI Innovation Challenge. At this week's Kick-Off Day in Edinburgh, Mike Oaten, CEO at TIKOS® introduced our AI Bias Detection and Mitigation proposition. Because TIKOS® accesses a model's internal decisions (not just its inputs and outputs), we can pinpoint the exact neurons and activation pathways producing a biased outcome, edit the model's weights to remedy it, and monitor in production. Algorithmic bias is acknowledged as a significant source of harm for consumers by firms, regulators and consumer groups - and getting this fixed now is a priority! We've validated our solution in the lab. The next step is field testing with the programme partner firms who are actively looking for a solution to this problem, including Baillie Gifford, Barclays, BlackRock, Equifax, EY, HSBC, Lloyds Banking Group, NatWest Group and PwC. Thanks to FinTech Scotland and the whole FRIL cohort for a great start to the programme. Build Trustworthy AI. #FRIL #FinTechScotland #ResponsibleAI #AIBias #FinancialServices

TIKOS® has been awarded a £140,000 Innovate UK Growth Catalyst grant to advance our AI Assurance platform for regulated industries. Over the next 12 months, the funding will help accelerate our core, patent-pending technologies and overall product roadmap. The award is further validation that current approaches to AI testing and monitoring fall short where AI enabled systems can impact people's health, safety or fundamental rights. Existing solutions observe model inputs and outputs. TIKOS® captures inference data from inside the model at run-time, replacing estimation with measurement. Exactly the hard evidence needed for regulated organisations to build and deploy compliant AI with confidence. We're grateful to Symvan Capital, whose equity investment was essential to unlock this grant under Innovate UK's investor-matched funding model. Thank you to Michael Theodosiou and the Symvan team for their ongoing support. 🙏 Build Trustworthy AI. #TrustworthyAI #AIAssurance #InnovateUK #DeepTech

TIKOS® just published results of a proof-of-technology project ran with Cisco supported by the UK's LASR (Laboratory for AI Security Research) and UK Government. The Challenge Multi-Agent-Systems (MAS) deployed in critical infrastructure must be precisely controlled. A compromised system can quickly threaten life, multi-million-pound assets, and hard-to-stop cascading failures. Subtle, sophisticated prompt injection attacks present a wide and fast evolving threat surface. Existing approaches operate at the input/output boundary, scanning for suspicious keywords, running output classifiers, maintaining blacklists. But sophisticated attackers don't use suspicious keywords. They use authority claims, technical jargon, emotional manipulation, and fake policy documents. It's a vocabulary arms race, and attackers always find new techniques. TIKOS® introduces a new approach. Instead of asking what did the model generate, we ask how did the model decide to generate it, by analysing the model's internal neural pathways in real time. Think of it as an MRI for an agent's reasoning. What We Discovered Regardless of how an attacker frames their request (fake credentials, gaslighting, emotional pressure, format tricks) the model's internal architecture responds in a similar (mathematical) way. A property identified during the project and which we call 'Intent Signatures'. This technique has the advantage of stopping zero-day attacks because identification operates at a higher level of abstraction (the intent of the input, not the implementation of that intent). The results (600 unseen test cases, across 5 distinct attack types): 🎯 100% detection rate ✅ Zero false positives ⚡ <35ms latency overhead Full write-up (including video): https://lnkd.in/ewQQ24sa Thank you to the teams at Cisco and LASR / Plexal / Plexal Cyber for a genuinely challenging and rewarding programme and for all the support they provided to our founders Mike Oaten and Dr Don Liyanage : Adele Every, Declan Hadley, Mustafa Mustafa, Bradley Rossi, Simon Dyke, Simon Lidgett, Sumeet Bhatia, Darren Lewis (CMgr FCMI), Patricia Jamelska, Derya Atlas.

We're launching the TIKOS® Research Programme. The programme gives academic and applied researchers access to the TIKOS® AI assurance platform to investigate problems at the intersection of AI robustness, safety, explainability, and regulatory compliance, in sectors where failure has real consequences - healthcare, defence, critical infrastructure. TIKOS® is the only platform that captures and analyses the internal operations of deep-learning models at inference time. That means researchers can study what a model actually did, not just what it output, at the point of a decision. This capability enables research that has not previously been possible at scale: causally-grounded explanations of individual decisions, real-time adversarial attack detection at the model internals level, and the generation of formal compliance evidence derived directly from inference data. We're opening the programme to researchers working in five sectors: - AI security - Defence and national security - Critical infrastructure - Healthcare and clinical AI - Financial services and regulatory compliance Accepted participants receive structured platform access, including API and SDK integration, Synapses Logger, Causality Traces, our case-based reasoning engine, and tooling to build a suite of profilers for bias, robustness, accuracy, and cybersecurity evaluation, etc. TIKOS® contributes up to £10,000 in licence, compute, and support costs per project. In return, participants commit to producing at least one output that advances the evidence base for trustworthy AI in regulated industries (a publication, conference paper, technical report, or substantive case study). Applications are now open. https://lnkd.in/euzdjV4u #TrustworthyAI #AISafety #AIResearch #ExplainableAI

🚀 We're Hiring: Technical Specialist | Tikos Technologies Ltd At Tikos Technologies, we're building the next generation of decision-reasoning, where high-speed distributed systems meet cutting-edge AI. And we need a rare kind of engineer to help lead that charge. This isn't your average technical role. We're looking for someone who can write Python or Go in the morning, debate PyTorch inference optimisation in the afternoon, and still have the vision to architect systems that are secure, scalable, and audit-ready. 🔧 What you'll own: → Architecture & delivery of distributed, large-scale data systems → Deployment of deep learning models (PyTorch) optimised for low latency → ISO27001 & SOC2 compliance, from access control to audit evidence → Secure Git pipelines, Docker deployments & 99.9% uptime targets → Mentoring a talented team with a "Security by Design" ethos 🛠️ Our stack: Python · Go · PyTorch · Gemini/Claude · GCP · Docker · Graph & Relational DBs Apply at: https://lnkd.in/e8AfPgMM If you want to discuss about the role contact Don Liyanage

We're #hiring a new Technical Specialist in Greater Bristol Area, United Kingdom. Apply today or share this post with your network.

Happy 125th Anniversary to the British Standards Institution (BSI)! 🎉 For over a century, BSI has been at the forefront of driving global progress evolving from standardizing steel in 1901 to pioneering the modern frameworks for digital trust, sustainability, and Artificial Intelligence today. As the world's first National Standards Body, their enduring legacy of raising standards continues to shape a safer, more reliable, and innovative world.   You can explore more about BSI’s incredible 125-year journey here: https://lnkd.in/ekf2saB2   At TIKOS®, we know that the foundation of building truly Trustworthy AI lies in rigorous assurance, transparency, and robust governance. That is why we carefully follow BSI standards and closely monitor BSI's vital contributions to the ISO AI standards (such as ISO/IEC 42001). These global benchmarks are central to the development of our core product strategies, ensuring that our AI evaluation and explainability tools meet the highest regulatory and ethical demands.   We are also incredibly proud to share that TIKOS® Co-founder, Dr. Don Liyanage, is actively contributing to these global standardisation efforts. Serving as a national-level expert on the BSI ART/1 (Artificial Intelligence) Standards Committee, Dr. Liyanage is working to bridge the critical gap between cutting-edge academic research and practical industrial application.   Here’s to BSI’s continued leadership in global excellence, and to our shared mission of building a future powered by safe, accountable, and standardised AI!   #BSI125 #TrustworthyAI #AIStandards #TikosTechnologies #ArtificialIntelligence #AIGovernance #ISO42001 #Innovation #TechLeadership #AIAssurance