Team Atlanta spent two years exploring how artificial intelligence (AI) could help identify and fix software security problems as part of the Defense Advanced Research Projects Agency (DARPA) AI Cyber Challenge (AIxCC).

After a grueling, world-class competition, the team won the $4 million first-place prize in August 2025. 

The team has since disbanded. Some members have moved into the corporate world, while others continue in academia. Despite these varied paths, Team Atlanta’s search for breakthrough AI cybersecurity solutions continues.

From AIxCC to Microsoft

Team Atlanta lead Taesoo Kim, principal security researcher Dongkwan Kim, and three other former team members joined Microsoft Security after the competition, where they continue to work on AI and cybersecurity in the Frontier Offensive Research & Generative Exploitation (FORGE) Lab.

The move has given the researchers an opportunity to apply concepts from AIxCC to real-world software. It has also exposed them to challenges that are difficult to replicate in a competition or research lab.

Dongkwan Kim said that team members continue to share what they learned through research papers, reports, software, and benchmarks. At Microsoft, these lessons are tested against real customer code.

This work led to MDASH, Microsoft's multi-agent, multi-model system for vulnerability discovery and remediation. The system uses more than 100 AI agents to search for vulnerabilities, validate potential findings, and help develop fixes.

Rather than asking a single AI model if a piece of code contains a vulnerability, MDASH can examine a problem using multiple agents. The system can also try to create an exploit to determine whether a potential vulnerability is real and reachable.

For Dongkwan Kim and the other former Team Atlanta members at Microsoft, moving from AIxCC to MDASH showed how research can look different when applied to real software.

“AIxCC was not an endpoint for Team Atlanta,” he said. “We continue sharing what we learned through technical reports, papers, open artifacts, and benchmarks for vulnerability discovery and patching.”

Working with customer code introduced new challenges for the Microsoft team. The researchers had to consider security boundaries, existing development processes, how findings would be reviewed, and how vulnerabilities would be prioritized for action.

These challenges created a new feedback loop between research and product development. The team took ideas from AIxCC, test them on real software, learn where they worked and where they fell short, and use those lessons to improve the system.

The experience also showed that AI does not replace the need for cybersecurity experts. Human researchers are still needed to provide context, decide how to break down difficult problems, validate findings, understand attack paths, and determine the potential impact of a vulnerability.

Taesoo Kim, a professor in the School of Cybersecurity and Privacy and vice president of security research at Microsoft, said the competition continues to influence the Microsoft team's work.

“Every single activity that we’ve done at Microsoft was influenced by AIxCC,” he said in a 2026 interview with Microsoft.

The work has continued to grow. In 2026, Microsoft introduced MAI-Cyber-1-Flash, an AI model built specifically for cybersecurity and designed to work with MDASH.

The model is designed to handle many of MDASH's tasks while allowing larger AI models to focus on harder problems. Microsoft reported that replacing 80% of the models used by MDASH with MAI-Cyber-1-Flash increased its performance on the CyberGym benchmark from 88.4% to 95.95%.

The benchmark includes more than 1,500 examples from 188 software projects. Microsoft also reported that the new setup can achieve similar performance at half the cost of the previous MDASH configuration.

“AIxCC has sparked cybersecurity innovation across the globe, from frontier AI companies to small academic labs,” said Taesoo Kim. “It is an incredibly exciting time to be working in AI-driven security engineering and research, as we see AI fundamentally reshape how we find and fix vulnerabilities at scale.”

Continuing the Research at Georgia Tech

The AIxCC experience also continues to shape cybersecurity research at Georgia Tech.

Jiho Kim, a Georgia Tech Ph.D. student who was part of Team Atlanta's research efforts, said one of the biggest lessons he’s brought back to campus was learning how AI and traditional security tools can work together.

One of these tools is fuzzing. Fuzzing tests software with large numbers of inputs to detect unexpected behavior or security problems. Large language models (LLMs), meanwhile, can reason about code and help guide the search.

During the AIxCC competition, the Team Atlanta combined these ideas in CtxFuzz, a system that uses an LLM to help guide fuzzing toward parts of software that may contain vulnerabilities.

The research was submitted to the ACM/IEEE International Conference on Automated Software Engineering, where it received the ACM Distinguished Paper Award.

“Since AIxCC, we’ve continued building on some of the core findings from our two-year journey with Team Atlanta,” Jiho Kim said.

“One of the biggest lessons for us was understanding how LLMs and traditional fuzzing can work together in a mutually beneficial way, using each to overcome the limitations of the other and ultimately find more and deeper vulnerabilities.”

The researchers also wanted to make the lessons from AIxCC useful to other cybersecurity researchers.

They created an Open-Source Cyber Reasoning System Framework (OSS-CRS) and CRSBench, which provides researchers and security professionals with ways to use, test, and build on technology developed through AIxCC.

“We’ve also been focused on making the results of AIxCC useful to the broader security community,” Kim said. “We created OSS-CRS and CRSBench so that researchers and practitioners can more easily use, reproduce, and build upon the capabilities and lessons from our AIxCC work.”

The team's research has also moved beyond identifying vulnerabilities. Researchers are studying how well AI models can understand and reason about computer code.

Jiho Kim and his lab mates in the Systems Software & Security Lab developed COIN, a benchmark designed to test how well LLMs can reason about code. The work has been accepted to the 2026 Conference on Empirical Methods in Natural Language Processing.

The Georgia Tech researchers are also developing additional benchmarks to study the various types of reasoning required for cybersecurity. They are using what they learn to create new datasets and methods for training AI models focused on cybersecurity.

“We’re using what we learn from these benchmarks to train cybersecurity-focused models through new datasets and training methodologies,” Jiho Kim said.

Photos by John Popham/College of Computing

News Contact

Seyedahmad “Ahmad” Rahimi, associate professor in Georgia Tech’s College of Lifetime Learning, has received a $1.6 million CAREER Award from the National Science Foundation (NSF). His interdisciplinary research explores how artificial intelligence can help middle school students build creativity and deepen their understanding of physics.   

The Faculty Early Career Development (CAREER) Program is NSF’s most prestigious honor for early-career faculty with the potential to become leaders in research and education. Rahimi, the College’s first tenure-track faculty member, is also the first to receive the honor.

“I’m very humbled to have received this CAREER Award because it represents an investment in my growth as a scholar exploring the intersection of artificial intelligence, learning sciences, psychometrics, and educational technology,” Rahimi said. “It’s very rewarding to see that my work has been recognized and to have this opportunity to investigate questions that are deeply important to me.”     

At the center of Rahimi’s project is Little C., a generative AI coach named for the concept of everyday creativity. The coach will assess students’ designs and offer feedback as they create, test, and revise physics-based simulations. Students will use the coach within Physics Playground, an existing digital sandbox where they design and test game levels to explore Newton’s laws of motion. By pairing creative exploration with physics concepts, Rahimi hopes students build STEM skills, gain confidence in trying something new, or build learner agency.

“What if we could have an entry point into STEM learning that looks less intimidating through the game environment and give students the opportunity to create, experiment, learn, and be creative?” Rahimi said. “Instead of treating creativity and learning as separate things, we want students to learn through creating and become more creative through learning; this is a win-win situation."

The project, which will take place primarily in after-school settings, will also explore how students engage with the AI coach during design activities and how that engagement influences creative output, physics learning, and AI literacy. 

“I’m trying to make students interested in physics,” he said. “Through this fun activity in the context of after-school programs, students can build this interest and feel self-efficacious about physics and creativity.”

During the project’s first year, Rahimi plans to engage educators and learners as co-designers to help ensure Little C. is as accessible, effective, and useful as possible. Their input will help shape the AI coach’s communication style, learning supports, and safeguards, all designed to support the responsible and appropriate use of AI in educational settings.  

This project builds on research underway in Rahimi’s CLASS Lab, or Creative Learning through AI and Stealth Assessment for STEM. The lab focuses on how emerging technologies can support learning, problem-solving, and creativity while keeping learners’ interests and abilities at the center of the experience. 

In fact, creativity has been a recurring theme throughout Rahimi’s career. He traces that interest back to fourth grade, when an art teacher praised the creativity he showed in a drawing. He said that experience helped shape his long-standing interest in how creativity develops and its role in learning.   

“We are living in an uncertain era and problems that we have never seen before will come our way, and we need people who have original thinking who can deal with those problems,” Rahimi said. “I believe creativity has been and certainly is one of the most important, if not the most important skill for human beings.”

Author: Joëlle Walls

Deleting a conversation with an artificial intelligence (AI) assistant may not erase everything the assistant remembers.

Georgia Tech researchers have developed a tool that can recover deleted conversations from AI assistants running on Android phones. The tool, called ORISA, can also rebuild the recovered information into a working AI session.

The findings show that deleting an AI conversation on a phone does not necessarily mean it removes every trace.

The research could give digital investigators a new way to recover information from phones. It also raises privacy concerns for people who use AI assistants that run locally on their devices.

Local AI assistants run directly on a phone rather than sending conversations to a cloud service. While the local assistant is running, information from conversations can remain in the phone’s working memory even after it is deleted.

ORISA searches this memory for pieces of deleted conversations.

In one experiment, researchers created a fictional AI assistant called “Project Note Keeper” using the ChatterUI Android app. They gave it 100 fictional facts about a project, then deleted the conversation.

ORISA recovered information from the deleted conversation and used it to create a new working session. When researchers asked the restored assistant to list the project notes it had been given, it returned 26 facts. The original session could access only 14 facts because the rest had fallen outside its active memory.

The experiment showed that researchers could recover information the AI had already stopped actively using.

“It can help establish what information someone shared with an AI assistant and what assistance they received,” said Ph.D. student Ken Xu, lead author of the study.

The researchers tested ORISA with 10 Android apps and three AI models. Across 60 test configurations, the tool recovered conversation information that remained in the phone’s memory.

The researchers say the information becomes harder to recover as the phone’s memory is overwritten by other activity.

The research was presented at the 2026 IEEE Symposium on Security and Privacy. The paper, titled Recovering and Rehosting Mobile Local LLM Conversations and Contexts via Memory Forensics, was authored by Ph.D. students Xu, David Oygenblik, Runze Zhang, and Mingxuan Yao; Research Scientist Muhammad Ibrahim; and Associate Professor Brendan Saltaformaggio. The work was done in Georgia Tech's Cyber Forensics Innovation Laboratory, a collaboration between the School of Cybersecurity and Privacy and the School of Electrical and Computer Engineering. 

News Contact

Georgia Tech researchers, educators, students, and industry leaders met at Exhibition Hall on Sept. 28 for the Tech AI Symposium to address the challenge of turning AI research into practical tools.

Two morning panels explored research collaboration and workforce adaptation, while an afternoon student poster session and a concurrent Tech AI Career Fair connected the discussions with research and career opportunities.

For Tim Brown, Georgia Tech’s academic program director for AI, helping people find the right expertise is a practical starting point. When people ask how to become involved in AI research and partnerships, he said, Tech AI can help connect their interests with researchers and projects.

“We get these inquiries every day,” Brown said. He encouraged students to share their résumés and areas of interest, and to seek out faculty whose research methods and fields of study align with their goals.

Tech AI’s broader purpose is to help industry, government, and other organizations move from research ideas to AI systems they can put to use. Part of the university-wide Georgia Tech AI ecosystem, Tech AI draws on expertise across the Institute’s colleges and schools, interdisciplinary research institutes, AI institutes, advanced computing infrastructure, and the Georgia Tech Research Institute (GTRI).

The "How Tech AI Collaborates" panel, moderated by Brown, examined what it takes to make those partnerships work, particularly when research depends on sensitive information or creates intellectual property.

Belinda Vogel, research engagement manager for the Georgia Tech Manufacturing Institute and the Renewable Bioproducts Institute, described plans to open advanced manufacturing capabilities to remote users in academia and industry. Such access could extend the reach of research facilities, but it also raises questions about how companies can benefit from AI while keeping their data private.

Kevin Dalmeijer, a senior research associate in Georgia Tech’s H. Milton Stewart School of Industrial and Systems Engineering and managing director of the NSF AI Institute for Advances in Optimization, described safeguards used in projects involving sensitive or controlled data. Those measures can restrict where information is stored, how students access it, and which AI tools they can use.

Dalmeijer also discussed internships that divide a student’s time between Georgia Tech research and work embedded within a company. Establishing clear boundaries at the outset, he said, helps distinguish academic research from company-specific implementations and clarifies expectations for data and intellectual property.

Brown brought Ajay Upadhyaya, Tech AI’s associate director for research administration, from the audience into the discussion to explain another approach. Upadhyaya described developing foundational models with synthetic data, allowing students to publish academic work, then defining separate ownership for work incorporating a company’s proprietary data. He emphasized the importance of clearly specifying those boundaries.

The panel also looked toward emerging computing capabilities. Jeffrey Young, a Georgia Tech principal research scientist working with the Partnership for an Advanced Computing Environment, discussed how AI could help researchers improve quantum systems and determine how best to map applications onto quantum hardware. His remarks pointed to a relationship in which AI can support the development of other computing technologies, even as researchers continue exploring their potential.

The second panel, “Workforce Adaptation in the Age of AI,” shifted the focus to the people learning and working with these tools. Moderator Leigh Hopkins, assistant director of Georgia Tech’s Center for Economic Development Research, described a workforce being asked to adapt while the technology itself rapidly changes.

David Joyner, Georgia Tech’s interim vice provost for AI in Education and executive director of its Online Master of Science in Computer Science program, said AI familiarity is moving toward becoming an expected professional skill. He compared that transition with an earlier era when applicants listed typing speed on their résumés before such capabilities became “baseline.”

People still need the knowledge and judgment to do the underlying work themselves, Joyner said. Learning to use AI must be paired with learning to evaluate its output and avoid overreliance on the technology.

Georgia Tech faculty are exploring that balance through classroom experimentation, from changing assignments to developing tools tailored to particular courses. Joyner described the Academic AI Playbook as a way to share what individual instructors tried, the challenges they encountered, and what others might learn from their experiences. The 2026 edition includes 67 faculty-authored articles and 20 lessons drawn from classroom experience.

Joyner also raised a longer-term question for employers and educators: If AI takes on tasks traditionally assigned to entry-level employees, how will those employees gain the experience needed to advance? He suggested that evaluating work, framing problems, and defining clear requirements will become increasingly important as people take responsibility for directing AI-assisted work.

Georgia Tech alum Brian Greene, an operations and supply-chain executive, brought the discussion back to organizational decision-making. Companies can become so focused on acquiring an AI tool, he said, that they overlook the business need it is supposed to address.

“What’s the problem you’re trying to solve?” Greene asked. He urged organizations to begin with that question and work backward to the capabilities and partnerships they need. For smaller businesses with limited resources, choosing a tool without a clear purpose can be especially costly.

Narayanan (KK) Krishnakumar, chief technology officer at Delta Air Lines, offered examples of AI at work in a large organization. He described systems that help optimize baggage routing and generative AI tools that summarize information for customer-service employees. Across those applications, he emphasized the continuing need to verify AI-generated information.

For both Krishnakumar and Greene, the qualities employers value remain the same as they have traditionally been: curiosity, problem-solving, collaboration, and an ability to deliver results. AI fluency adds to those capabilities while making sound judgments about data and answers all the more necessary.

The afternoon poster session demonstrated how widely those questions now reach. Student projects explored trustworthy optimization, AI agents and responsible design alongside research on education, city planning and biomedical diagnostics. Other work addressed materials discovery and sustainable material selection, quantum chemistry, signal analysis, and AI that could support aging in place.

A well-attended career fair ran simultaneously with the symposium. In a large auditorium filled with company booths, students lined up to meet employers as both sides sought to connect research with potential positive industry outcomes.

News Contact

News Contact

News Contact

News Contact