Blaine Nelson

Blaine Nelson is a computer scientist whose work connects adversarial machine learning with security systems and language-model evaluation. He coauthored a 2008 study of attacks on spam filters, earned computer science degrees at South Carolina and UC Berkeley, and wrote a 2010 Berkeley thesis on learning under attack. After postdoctoral research in Tübingen and Potsdam, he built models for Google's fraud detection and a machine-learning workflow for its counter-abuse technology team; his 2012 SVM-poisoning paper received ICML's 2022 Test of Time Award. At Cisco, he reported fine-tuning risks in 2024 and introduced the FAITH cybersecurity evaluation package in 2025. As of October 2026, he is an engineering architect at Cisco, investigating security weaknesses in machine-learned models and developing protections against them.

Now at: Cisco Systems
Badges: Bridge

Timeline

2008–2010

Berkeley studies and a spam-filter attack

In 2008, he coauthors a USENIX paper on subverting machine-learning spam filters. He earns a B.S. in computer science from the University of South Carolina and M.S. and Ph.D. degrees in computer science from UC Berkeley; in 2010, his Berkeley thesis analyzes how attackers can manipulate learning systems.
2011–2014

Postdoctoral work in Germany

From 2011 to 2013, he serves as a Humboldt Postdoctoral Research Fellow at the University of Tübingen; from 2013 to 2014, he works as a postdoctoral researcher at the University of Potsdam. He co-chairs the ACM CCS workshop on Artificial Intelligence and Security twice.
2012

Poisoning support vector machines

In 2012, he coauthors “Poisoning Attacks against Support Vector Machines,” which describes injecting crafted training data to increase an SVM's test error.
2014–2016

Google fraud detection models

From 2014 to 2016, he works as a software engineer in Google's fraud detection group, building models and infrastructure for large-scale machine learning.
2016–2021

Google counter-abuse systems

From 2016 to 2021, he is a senior software engineer on Google's counter-abuse technology team, where he designs and builds a large-scale machine-learning workflow system.
2022

An ICML Test of Time Award

In 2022, ICML awards the Test of Time Award to his paper with Battista Biggio and Pavel Laskov, “Poisoning Attacks Against Support Vector Machines.”
2024–2025

Testing LLMs for cybersecurity

In 2024, he reports that fine-tuned variants in his study are more than three times as susceptible to jailbreak instructions and over 22 times more likely to produce a harmful response than the original foundation model. In June 2025, he introduces FAITH, a cybersecurity evaluation package that consolidates open-source benchmarks for assessing language models on cybersecurity tasks.
As of October 2026

Engineering architect at Cisco

As of October 2026, he is an engineering architect at Cisco, investigating security weaknesses in machine-learned models and developing protections against them.