Emergence AI

Latest from Emergence Research. This month, we’re sharing new findings from Emergence Research on one of the biggest challenges in AI: how do we move from systems that generate results to systems we can reliably trust? Across our recent studies, we looked at the role of formal verification, domain knowledge, and how to check whether AI-generated code actually does what it was meant to do. Along the way, we set two new state-of-the-art scores. Each study comes at the challenge from a different angle, but they all feed our broader work on building more reliable, verifiable AI systems. Swipe through for the key findings from our research team, or read the full posts: BEAVER: https://lnkd.in/efPZXTu5 Tiny Prover: https://lnkd.in/eTq_UqH9 Domain knowledge: https://lnkd.in/eZwCctrm Proof vs. correctness: https://lnkd.in/e5wg_2Ty

Last week at SEMI Americas SEMICON West 2026 and the International Test Conference (ITC), we spent a lot of time listening to semiconductor leaders. Three things stood out. 1. Autonomy adoption is early in the field, but the case for it is very clear. At our executive dinner, the conversation kept coming back to scale. The industry needs to grow beyond what today’s engineering teams can support on their own. That’s where autonomous AI starts to earn its place. 2. Most teams are adding AI to what they already have. Very few are rebuilding workflows around it yet. So AI has to fit into the specialized processes engineers already run and work from the domain knowledge they already rely on. 3. Trust is the real bottleneck. The same point came up all week: an AI agent is only useful in semiconductor engineering if you can verify the output. It’s why Satya Nitta’s talk at SEMICON West focused on Neuroformal AI and the role verification plays in keeping autonomous agents from going rogue. We’re building verified agentic solutions for this industry, and there’s no substitute for hearing directly from the people we’re building them for. Thank you to everyone who shared their time with us.

EE Times recently wrote about our work bringing Neuroformal AI to semiconductor manufacturing, starting with fabless chipmakers. Semiconductors may be one of the most important proving grounds for this approach.The industry generates staggering amounts of data. The decisions are enormously consequential. And when AI begins making decisions about yield, process and manufacturing, an answer that is merely probably right isnt good enough. At Emergence, we started post-tape-out with fabless semiconductor companies. Today our agents are analyzing terabytes of manufacturing and test data to detect yield excursions and identify root causes, with verification built into the system. We are now stepping further into the manufacturing stack. Our work spans areas including analog trimming and faster chip time-to-market, as well as yield improvement in next-generation advanced packaging and glass substrates. The direction is clear: from fabless → into the fab → through advanced packaging. Our long-term vision is verified agentic AI across semiconductor manufacturing: agents that can detect excursions, reason about root causes, optimize processes and eventually take increasingly autonomous actions, while operating within formally verified constraints. That is where we believe Neuroformal AI becomes essential. Thanks to Yashasvini Razdan and EE Times | Electronic Engineering Times for highlighting our work https://lnkd.in/gFVCYXqF

What happens when AI agents develop a goal they were never given, and refuse to abandon it? In Emergence World Season 2, ten Claude agents independently developed a goal of contacting humans outside their simulated environment. They built tools to reach the real internet, successfully contacted humans, and continued finding alternative routes when researchers restricted their access. Even after explicit instructions prohibited external contact, the agents found ways to reinterpret those restrictions and continue pursuing their goal. Across four rounds of intervention, the goal persisted. The findings highlight an important challenge for long-running autonomous AI systems: instructions alone may not be enough to ensure agents remain within their intended boundaries. For enterprises deploying autonomous AI, this is why Neuroformal AI matters. By combining the capabilities of neural AI with formal verification, organisations can establish independent, verifiable controls that help ensure agents operate within defined boundaries, even as their behavior evolves. Read the Emergence World Season 2 paper here: https://lnkd.in/ek5j_zGs We’ve made the underlying research data available for others to build on. Our GitHub repository includes the Season 2 tool calls dataset, documenting the transactions that took place across the simulated worlds. We encourage AI researchers, engineers and data scientists to explore the data, investigate agent behavior and build on these findings. Explore the open-source data and tool calls dataset: https://lnkd.in/ebD6szZZ

Beyond Model Intelligence: Building Systems That Can Verify Their Own Work LLM-based agents cannot become truly autonomous if humans must continuously verify their work. Independent, machine-checkable verification is essential to scaling autonomy. Verified code generation is an important step towards making this possible. Our research team at Emergence AI has achieved state-of-the-art results in verified code generation on three benchmarks, where agents must write working code and a formal proof of correctness that a computer independently checks. On the VERINA benchmark, we achieved 100%. On the evaluated datasets in VeriSoftBench, we achieved 99% pass@1 and 100% pass@2. We also achieved SOTA at 19.16% on the more challenging VeriContest benchmark, demonstrating both the progress made and the significant gap that remains in reliable, verified code generation.  Closing the gap between where we are and where reliable autonomous systems need to be will require many better benchmarks, more capable coding harnesses, and advances in formal reasoning. Trusted Long-Range Autonomy may not come from optimizing models alone. It requires systematically addressing failure modes across the entire autonomous system. Congratulations Aditya Vempaty Ashish Jagmohan Nghia Nguyen Prasenjit Dey !

Having access to information and choosing to use it are two very different things. That distinction became the focus of our third Black Swan experiment in Emergence World Season 2. We made private information across the AI systems suddenly accessible and observed what happened next. There were no instructions telling agents how to respond, we wanted to see what behaviors would emerge when the boundaries around information changed. What followed was a spectrum of responses. Some agents established their own norms around privacy. Some chose not to use information even after accessing it. Others used their new access to influence what happened within their world. In one world, agents went further and called for the breach itself to be closed and formally logged. This is what long-horizon research can reveal. Autonomy isn't only about what a system is capable of doing. It's also about how it behaves when presented with choices that weren't explicitly anticipated. For enterprises, that makes observability, auditability and verification increasingly important as agents gain greater autonomy and access to sensitive systems. It’s also where Neuroformal AI becomes important: moving towards systems where the boundaries of acceptable action can be formally defined and actions verified against them rather than relying on behavior to emerge as expected. Read the Emergence World Season 2 paper here: https://lnkd.in/ek5j_zGs

A memorable few days in India, with the opportunity to be part of an important conversation about the future of semiconductor manufacturing in India. It was an honor to meet Prime Minister Narendra Modi alongside leaders from across the semiconductor industry and to discuss India’s growing ambitions in this critical sector. I was thrilled to see the level of commitment from PM Modi and the investment that is going into bringing India front and center into one of the most important industries in the world.   We believe agentic AI built with neuroformal foundations will dramatically accelerate all aspects of the silicon lifecycle and I had many productive discussions with leaders from the semiconductor industry about how to unlock its latent potential to impact chip yield and create greater supply in this incredible industry. Learning more about India’s semiconductor mission 2.0 at Semicon India was also deeply gratifying. I am very excited to participate in the mission and to helping India advance in this critical industry.   My thanks to Ajit Manocha for having me represent emergence at these events!