When a deal closes, the press release talks about product and talent. The security team inherits something else.
Raj Handa, Senior Cybersecurity Engineer in offensive security at Workday, leads its pentest program and runs its bug bounty. When I asked him on Scale To Zero what a typical day looks like, acquisitions came up right away. Like many SaaS companies, Workday has been acquiring for the past few years.
“So we acquire a company, inherit their infrastructure, realize that they sometimes have no security program, and we have to triage what gets hammered before it even touches a customer.”
This post covers how Raj brings an acquisition into a bug bounty and pentest program, and the longer-term threat he thinks those acquisitions, especially AI ones, will bring with them.
The Cardinal Rule
“You never dump an acquired company straight into your public bug bounty program on day one.”
When you buy a company, Raj said, you inherit their infrastructure, their technical debt, and their forgotten staging servers. You are also onboarding an engineering team that may never have operated under the scrutiny of a public bug bounty.
Flip that switch immediately and three things happen:
-
Researchers flood your inbox with low-hanging fruit: unpatched frameworks, configuration oversights, and more.
-
You blow through your bounty budget within a few days.
-
Most importantly, you crush the morale of the acquired engineering team. Nobody wants to handle a pile of security issues on day one of their acquisition.
The Three-Phase Runway
To bring an acquisition into scope safely, Raj uses what he called “a disciplined three-phase runway.”
Phase 1: Internal Offensive Discovery
Before any external researcher touches the acquired surface, the internal team runs attack surface discovery and targeted penetration testing. The priority isn’t only testing the web app. It is mapping trust boundaries:
-
How does the acquired network connect to the corporate network?
-
What identities bridge the two?
This is also where you find and fix the embarrassing basics internally. Raj’s list: open storage buckets, missing MFA, outdated endpoints, and default credentials.
If you want a deeper look at the identity side of this step, our post on access handoffs when a cloud account changes hands covers the access that tends to outlive a transition.
Phase 2: A Private, Curated Bug Bounty
Next comes a time-bound, private, invite-only program. Raj gave HackerOne’s child programs as one example (and was clear he wasn’t marketing for them). His version runs for 30 days with handpicked, high-signal researchers who specialize in deep business logic and architectural bypasses. You give them the scope and let them hammer the environment for a few weeks.
One thing I’d add from our conversation: this phase depends on the researcher relationships you have already built. If you haven’t built them, a private program gets average researchers, and likely more AI-generated noise. We cover that in AI Slop in Bug Bounty.
Phase 3: Graduate to Public Scope
Only after phases one and two do you merge the acquired assets into the main public program. Raj’s exit criteria are specific: the acquired team’s mean time to remediate has stabilized, and their regression testing is tied into your enterprise CI/CD pipelines.
The Threat Coming With the Next Acquisitions
Sai Santosh Vernekar sent us a question that fits right here. With so many AI startups being built, and companies like Workday acquiring, where does Raj see the biggest cybersecurity threats emerging over the next three to five years?
His answer wasn’t deepfakes or AI-generated phishing.
“The real sleeper threat over the next three to five years is actually the silent breakdown of machine-to-machine trust in an autonomous agent ecosystem.”
Humans Are Still the Safety Brake
Today, you ask a model a question, it gives you text, and a human decides whether to accept, discard, or edit it. Raj expects that human buffer to be gone in three to five years. We are moving aggressively toward agentic workflows, where agents hold their own API keys and write and deploy their own code. He described agents negotiating vendor contracts, triggering financial transactions, and talking directly to agents belonging to other enterprises.
Deterministic Identity Meets a Non-Deterministic Agent
“Enterprise security is built on deterministic identity.”
We have spent decades building IAM around rigid rules: if you hold this certificate or token, you may run this query or call this API. Raj argued agents break that model.
“The agent is non-deterministic. Its behavior is governed by natural language and probabilistic reasoning and not really immutable logic gates.”
So an untrusted input, such as an unreviewed PR, an email with an adversarial instruction, or an indirect prompt injection, can hijack the agent’s intent. And, as Raj put it, “hence they have hijacked the entire agent-to-agent workflow.” He said this has already started showing up.
For practical background on these risks, see why AI coding agents become an identity and credential risk and our breakdown of an AGENTS.md prompt injection hijack.
Why These Two Topics Belong Together
This connection is mine, not Raj’s. Phase 1 asks which identities bridge the acquired company to yours. When the company you acquire is AI-first, some of those identities will be agents with their own keys. Mapping them before go-live is the same trust-boundary work, applied to the threat Raj expects to grow.
M&A Security Onboarding Checklist
-
No acquired assets in the public bug bounty on day one.
-
Internal attack surface discovery and targeted pentest completed first.
-
Trust boundaries mapped: network connections and identities bridging the two companies.
-
Basics fixed: open storage buckets, missing MFA, outdated endpoints, default credentials.
-
Time-bound private program (Raj uses 30 days) with handpicked, high-signal researchers.
-
Public scope only after mean time to remediate stabilizes and regression tests run in enterprise CI/CD.
-
Agent and service identities in the acquired environment inventoried, with their keys and permissions (my addition).
-
Agent workflows reviewed for untrusted inputs like PRs, emails, and fetched content (my addition, based on Raj’s threat description).
Listen to the Full Episode
Thanks to Sai Santosh Vernekar for the question. Raj also covers extension confusion in AI editors, pentesting as a portfolio, and how to never pay for the same bug twice. The extension confusion part has its own write-up: Extension Confusion: The IDE Supply Chain Risk Hiding in Your AI Editor.
-
Watch on YouTube: Pentesting, Bug Bounties & The IDE Supply Chain Blind Spot ft. Raj Handa
-
Listen on Apple Podcasts or Spotify
-
Join the conversation on LinkedIn
Scale To Zero is the security podcast we run at Cloudanix. If you have onboarded an acquisition into your security program, I would love to hear what you learned.
People Also Read
- Extension Confusion: The IDE Supply Chain Risk Hiding in Your AI Editor
- AI Slop in Bug Bounty: How to Keep Your Program From Becoming a Spam Folder
- How Often Should You Pentest? Treat Penetration Testing as a Portfolio
- Never Pay for the Same Bug Twice: Vulnerability Prioritization Beyond CVSS
- Why AI Coding Agents Become an Identity and Credential Risk