AI Security Research Lead

Belsoftsolutions; United States; Remote; USD 120,000 - 170,000; Full time

Job summary

Belsoftsolutions is seeking a lead researcher to drive exploration of how AI apps and agents fail in practice, focusing on weird inputs, prompt tricks, workflow abuse, and misuse patterns. You will manage a small team, maintain rigorous documentation, and translate findings into actionable patterns for engineers.

You will build trackers in spreadsheets, collaborate with engineers across time zones, and own quality improvements across testing cycles to ensure robust, secure AI products.

Qualifications

• Experience leading a team or operation in research, QA, manual testing, UAT, or a similarly structured delivery function.
• Strong hands with Excel or Sheets: build and run a real tracker, not just fill one in.
• Adversarial mindset: enjoy finding inputs/workflows that break systems and document it.
• Clear writing: describe a failure precisely so an engineer can reproduce it on the first try.
• Comfortable with a distributed, international team across time zones and time shifts.

Responsibilities

• Lead research into how AI apps and agents fail in practice: weird inputs, prompt tricks, workflow abuse, and misuse engineers might not think to try.
• Run manual security testing at the behavior level: what an agent can be tricked into doing, revealing, or acting on it shouldn’t
• Build and maintain test trackers, checklists, and findings logs in Excel or Sheets that stay usable as the team and product grow
• Turn recurring failure patterns into standing checks that feed DeployReady and client security audits
• Write up findings so engineers can reproduce and fix them without a translation layer: exact steps, input, and result
• Own quality improvement across testing cycles: track what keeps breaking, what got fixed, and what’s still open
• Communicate daily with engineers and other team leads, including across time zones, to prioritize and close out findings
• Lead and grow a small research/testing team over time, setting priorities and reviewing their work
• Keep researching how people misuse or break AI products, bring new tricks into the process

Skills

Team leadership

Excel/Sheets

Adversarial mindset

Clear writing

Distributed team

Education

Criminology or Psychology background

Tools

Excel/Sheets

Job description

Someone has to sit down and actually try to break the AI products we and our clients ship: feed them the weird input, push the workflow somewhere it shouldn't go, see what an agent can be talked into doing or exposing. You'll lead that research and run it as a disciplined operation rather than ad-hoc poking: tracked in Excel, documented precisely, and turned into patterns our engineers can design against and that feed DeployReady, our open-source scanner. You'll work closely with engineers and other leads across an international, remote team, translating what you find into something they can act on without back-and-forth, and you'll own quality improvement over time: what keeps breaking, what got fixed, what's still open. The job is research, judgment, and operational rigor; the engineers own the code.

What you'll do

• Lead research into how AI apps and agents fail in practice: weird inputs, prompt tricks, workflow abuse, and misuse an engineer might not think to try
• Run manual security testing at the behavior level: what an agent can be tricked into doing, revealing, or acting on that it shouldn't
• Build and maintain test trackers, checklists, and findings logs in Excel or Sheets that stay usable as the team and product grow
• Turn recurring failure patterns into standing checks that feed DeployReady and our client security audits
• Write up findings so engineers can reproduce and fix them without a translation layer: exact steps, exact input, exact result
• Own quality improvement across testing cycles: track what keeps breaking, what got fixed, and what's still open
• Communicate daily with engineers and other team leads, including across time zones, to prioritize and close out findings
• Lead and grow a small research and testing team over time, setting priorities and reviewing their work
• Keep researching how people actually misuse or break AI products, in the wild and in your own testing, and bring new tricks into the process

What you bring

• Experience leading a team or an operation, ideally in research, QA, manual testing, UAT, or a similarly structured delivery function
• Strong hands with Excel or Sheets: able to build and run a real tracker, not just fill one in
• A sharp, adversarial mindset: you enjoy finding the input or workflow that breaks something, and you're disciplined about documenting it
• Clear writing: you can describe a failure precisely enough that an engineer can reproduce it on the first try
• Comfortable working with a distributed, international team across time zones and time-shifted communication
• Genuine depth of interest in AI products and how people use, misuse, and break them
• Patient and consistent: this kind of research lives or dies on rigor, not on one clever find

Nice to have

• Experience testing chatbots, AI agents, or other AI-powered products, even informally
• Background in QA, red-teaming, trust & safety, content moderation, or incident/complaint handling
• A degree or research background that trained you to think about behavior, failure, or misuse, such as criminology or psychology
• Enough familiarity with how AI apps are built to ask engineers sharp, specific questions