IBM Research

Four stories you shouldn’t miss from IBM Research this week: 🤖 How hyperautomation is redefining work — Nicholas (Nick) Fuller explains the rise of intelligent AI agents. 🧬 How quantum computing is accelerating drug discovery, from molecular simulation to clinical trial optimization. 🔬 How a new photo-Hall effect method advances a 200-year-old pursuit in semiconductor physics. ⚛️ How IBM Quantum’s Loon and Nighthawk processors are bringing fault-tolerant quantum computing closer to reality. Read the full stories in our latest Circuit Breaker ⤵️

From the moon landing to advancing Earth science, IBM and NASA - National Aeronautics and Space Administration have always pushed the frontiers of discovery. 🚀 Today, we're pioneering open-source, geospatial AI models to accelerate climate insights and support a resilient future. In the latest IBM Smart Talks episode, IBM Research leader Dr. Juan Bernabe Moreno joins Malcolm Gladwell to discuss how this shared legacy is reaching new heights—and how planetary data is being democratized for global impact. Watch the full episode here: https://ibm.co/604189yVV

Why Do Enterprise Agents Fail? In a collaboration between IBM Research and UC Berkeley, we studied how agentic LLM systems break in real-world IT automation, for tasks involving incident triage, logs/metrics queries, and Kubernetes actions in long-horizon tool loops. Benchmarks typically reduce performance to a single number, telling you whether an agent failed but never why. To solve this black-box problem, we applied MAST (Multi-Agent System Failure Taxonomy), an emerging practice for diagnosing agentic reliability. By leveraging MAST to analyze ITBench—the industry benchmark for SRE, Security, and FinOps automation—we turned raw execution traces into structured failure signatures, revealing exactly what broke and how to fix it. We annotated 310 ITBench SRE traces across three distinct model classes: Gemini-3-Flash, Kimi-K2, and GPT-OSS-120B. Key Findings: - Frontier models like Gemini-3-Flash fail cleanly (2.6 failure modes/trace), typically hitting isolated bottlenecks like verification. Large open models like GPT-OSS-120B suffer from cascading failure modes (5.3 failure modes/trace). -A single reasoning mismatch early in the run poisons the context, leading to compounding hallucinations. - Across all models, the strongest predictor of failure is FM-3.3 (Incorrect Verification). Agents consistently "declare victory" without checking ground truth.  - Kimi-K2 struggles to recognize when a task is done. It exhibits a massive spike in Premature Termination (+46%) and Unaware of Termination Conditions (+43%), often quitting just before solving the problem or looping indefinitely. Takeaways from our analysis when building agents: - For Frontier Models like Gemini: Externalize Verification. Never let the LLM grade its own homework. Require hard tool evidence before exit. - Put termination + loop control outside the model: Termination issues are common killers (FM-1.5). Add explicit stop conditions + loop detectors for repeated tool calls/actions or implement Finite State Machines. - Force clarify-or-read-only when inputs are ambiguous: Clarification failures (FM-2.2) are a major failure driver for smaller models. Make ambiguity a first-class branch in your agent graph. If you’re building agents for enterprise IT workflows, this is the kind of evaluation you want: not just “did it pass?”, but “what broke, where, and what intervention is most leverageable?” Many thanks to the collaborators in this work: Saurabh Jha, Melissa Pan, Rohan Arora, Ayhan Sebin, Ion Stoica Resources: • Blog post: https://lnkd.in/gGAmfmjS • ITBench paper: https://lnkd.in/gN5QhjzA • ITBench code: https://lnkd.in/gu8Nkygt • MAST paper: https://lnkd.in/gWH4ha_Y  • MAST code: https://lnkd.in/gAs59keC #Agents #LLM #ReliabilityEngineering #SRE #Kubernetes #Observability #DevOps #AI

Our mission at IBM is clear: to build the world’s first large-scale, fault-tolerant quantum computer before the end of this decade. This is core to realizing our vision of the future of computing, one in which quantum computers, AI, and high-performance computing come together to solve problems that no single paradigm can solve alone.     The U.S. must act to remain at the forefront in quantum technology and today’s reintroduction of the National Quantum Initiative Reauthorization Act is a timely step to ensure the United States maintains its lead in quantum innovation. Our nation’s success will depend on strong coordination across research, workforce development, and public-private partnerships. That’s why we encourage Congress to act swiftly to strengthen the National Quantum Initiative to accelerate the real-world benefits of quantum computing – from drug discovery and materials to making our critical infrastructure more resilient.      At IBM, we’re committed to partnering with government, academia, and industry to translate quantum innovation into practical applications that will benefit all Americans.

Before the inbox fills up, let’s rewind 2025. Last year delivered transformative progress across AI, quantum computing, and emerging technologies at IBM. As we turn the page to a new year, we’re reflecting on the breakthroughs, firsts, and fascinating research that defined the past year across IBM Research and beyond. Catch up on ten of our favorite stories from 2025. Read the full article below.

Our vision for the future of computing is focused on integrating high-performance computing, quantum, and AI to tackle problems no single technology can solve alone – something we call quantum-centric supercomputing. This work goes hand-in-hand with the goals of the U.S. Department of Energy (DOE)'s Genesis Mission. Through IBM Quantum Innovation Center partnerships with DOE national laboratories, such as Lawrence Berkeley, Oak Ridge, and Los Alamos; and collaborations with four of the five DOE Quantum Information Science Research Centers, we are deeply embedded into the nation's quantum science ecosystem – providing a strong foundation for advancing the Genesis Mission. Together, we will pursue new algorithmic breakthroughs and explore how AI and quantum computing can complement each other to maintain and grow American leadership in advanced computing and scientific discovery. Read more on the Genesis Mission ➡️ https://ibm.co/6044BxiYm

What if the key to smarter, more efficient LLMs isn't bigger models—but better memory? IBM researchers just unveiled a new variation of state space models that restores a fundamental skill that transformers struggle with: state tracking, the basis for logical reasoning, sequential understanding, and even basic counting tasks. This breakthrough, PD-SSM, shows significant gains on parity, time-ordered event prediction, and long-sequence tasks—opening new doors for more capable and cost-efficient AI, including code generation. Explore how this model could reshape the future of LLM architecture → https://ibm.co/6046BQdJA

CUGA is now live on Hugging Face Spaces. This open-source generalist agent is built for complex, real-world workflows and enterprise experimentation, offering: ✅ High-performing generalist reasoning ✅ Reasoning modes to balance speed ✅ Multi-tool orchestration ✅ Integration with Langflow and Groq Explore how CUGA brings a new level of flexibility and openness to AI agent building, built in partnership with Langflow and Groq: https://ibm.co/6046Btzda

This week at IBM Research… 🚀 Granite 3.3 set a new benchmark for AI transparency, outperforming peers by 23 points. 📈 Our iSWE-Agent for Java now leads the Multi-SWE-Bench, tackling real‑world developer challenges. 🌐 Quantum experts Jerry Chow & Oliver Dial answered big questions about fault‑tolerant quantum computing. Read the LinkedIn article for more updates and deeper insights ⬇️

Transparency is built on details. IBM Granite earned the top spot on the Stanford Foundation Model Transparency Index by disclosing more information about our models, from training data sources to the resources used to build them. Released under a standard Apache 2.0 license, Granite is the first open model family to achieve ISO 42001 certification and is cryptographically signed, confirming adherence to globally recognized best practices for security, governance, and transparency. Discover why Granite is among the most transparent enterprise models ever developed: https://ibm.co/6043BoCuN