Feral Systems

A CIO advisory practice founded by Logan Browne — Amazon security leader, CISSP, originator of the Feral AI framework.

Feral AI — when every employee becomes a builder.

Feral Systems Wolf

Feral AI is the working software your employees are building with AI — most of it invisible to IT. AI has put builder power in the hands of everyone in your organization. Sales ops are shipping deal analyzers. Marketers are publishing custom GPTs. Finance is writing Copilot scripts that touch production data. We help CIOs turn that wave into governed advantage instead of a hidden liability.

Schedule an Advisory Briefing
↓

What is Feral AI?

The same emergence pattern we've studied for years — workers building their own solutions when formal IT falls short — now turbocharged by generative AI. Every non-builder is a builder.

The Non-Builder Builder

For decades, "feral" information systems looked like spreadsheets, Access databases, and personal scripts — humble tools that emerged because the official systems didn't fit. AI just collapsed the cost of building. A salesperson can now stand up a working deal analyzer in an afternoon. A marketer can ship a custom GPT to thousands of users. A finance analyst can wire an LLM into a recurring workflow that touches the general ledger.

Feral AI is not Shadow IT 2.0. Shadow IT was unauthorized software adoption. Feral AI is unauthorized software creation — at the speed and scale of every motivated employee in your org.

Our Position

Feral AI is not a problem to eliminate. It is a signal — the clearest signal you'll ever get about what your business actually needs, where your formal systems are failing, and which of your people are ready to build the next thing. Our job is to help CIOs see the unseen systems, channel it toward sanctioned platforms, and scale the winners.

Adaptive Information Systems

What Feral AI Looks Like

📊 The Deal Analyzer

A sales ops manager builds a Claude-powered tool that reads inbound RFPs, scores fit, and drafts response outlines. It spreads across the team in a week. Nobody on IT knows it exists.

🤖 The Custom GPT Library

Marketing publishes a dozen custom GPTs — brand voice generators, brief writers, competitor summarizers — as "tools" for the whole team. They handle confidential roadmap data on every prompt.

💼 The Copilot Workflow

A finance analyst uses Copilot to write a script that reconciles vendor invoices against the GL each Monday. It runs from her laptop. The close now depends on it.

📨 The Triage Agent

A support lead deploys a browser-based agent to read tickets, classify severity, and auto-respond to common questions. Customer data leaves the official help desk on every call.

🔗 The LLM-Glued Workflow

Ops connects Zapier + an LLM to summarize daily reports into Slack. Six months later, it's how leadership reads the business. The prompt template is in a Notion page only two people can find.

🧠 The Prompt Playbook

A consulting team builds a shared prompt library that produces client deliverables. It's now the firm's secret weapon — and lives in a shared Google Doc with no version control.

Why Feral AI Is Different

⚡ Build cost collapsed: What used to take an engineering sprint now takes an afternoon of prompting. Anyone with a problem and a chat window can ship.
🫥 It hides inside sanctioned apps: Unlike Shadow IT, Feral AI lives inside tools you already approved — Office, Google Workspace, the browser. Network controls and SaaS inventories won't see it.
🧬 Data leaves and doesn't come back: A pasted contract can be absorbed into model weights you don't control and can't request deletion from.
🤝 Identity sprawl: Each employee has their own AI accounts on their own credit cards. IAM, audit, and offboarding all break quietly.
🤖 Agents take actions: The newest wave doesn't just answer questions — it sends emails, books meetings, edits records, and writes code. Feral AI now operates on the business.
🚀 The upside is real: The same forces that create the risk also unlock genuine productivity. Banning rarely works. Channeling does.

The Stakes

Two forces are squeezing every CIO at once: workers building faster than defenders can secure, and attackers finding vulnerabilities faster than anyone can patch. Feral AI sits in the middle.

🔧 The Builder Side

  • 45% of AI-generated code samples introduce OWASP Top 10 vulnerabilities (Veracode).
  • 10× the rate of security findings from AI-assisted developers, who also commit 3–4× faster (industry research).
  • 107% year-over-year jump in vulnerabilities per codebase, attributed to AI-accelerated code creation (Black Duck OSSRA).
  • 62% of developers rely on AI suggestions for dependency selection — direct supply chain exposure.
  • 1 in 5 enterprise security breaches now involve AI-generated code.

⚔️ The Attacker Side

  • 2,000+ previously unknown vulnerabilities surfaced by a single AI system in 7 weeks (Anthropic Mythos).
  • 70%+ exploit development success rate with modern AI — up from under 1% with prior models.
  • 181 working exploits against Firefox's JS engine in one evaluation, where the predecessor produced 2.
  • 263% increase in CVE submissions from 2020–2025; NIST shifted the NVD to triage in April 2026.
  • 40% year-over-year rise in supply chain attacks in 2025.

The asymmetry: Your business is building faster than ever with AI, while adversaries are finding and exploiting faster than ever with AI. The gap is governance. When sanctioned alternatives exist, unauthorized AI use drops 89% — channeling, not banning, is what closes it.

The Feral AI Lifecycle

Stage 1: The Gap

A formal system can't do what the business needs today. Friction builds. Someone opens a chat window.

📉 Frustration meets a new shortcut

Stage 2: The Prompt

One employee solves the problem with an LLM, a custom GPT, or a Copilot script. Working software, made in hours.

💡 Non-builder ships a tool

Stage 3: The Spread

The tool proves valuable. Colleagues ask for the prompt, the link, the agent. It spreads — across teams, across data.

📈 Viral adoption inside sanctioned apps

Stage 4: The Dependency

A real business process now relies on it. Hallucinations and silent drift become operational risk. Data lineage gets murky.

⚠️ Mission-critical but ungoverned

Stage 5: Crisis or Channel

Either: a leak, a bad output, or an audit finding triggers a fire drill — or you channel it onto a sanctioned platform and scale the winner.

🔥 Decision point: channel or crisis

Not All Feral AI Is Equal

✅ Innovation to Scale

  • Solves a real, measurable problem
  • Uses sanctioned models and accounts
  • Documented prompts and inputs
  • Creator is responsive and visible
  • Limited sensitive-data exposure
  • Pattern is reusable across teams

→ Formalize, share, and scale

⚠️ Workaround Under Pressure

  • Solves a real problem with real risk
  • Mixed personal + work accounts
  • Sensitive data passes through prompts
  • Single-person dependency
  • Could break or drift without notice

→ Channel to a sanctioned path with a transition plan

❌ Hard Stop

  • Regulated data leaving the perimeter
  • Autonomous agents acting on production systems
  • Customer PII in consumer AI tools
  • No audit trail or human-in-the-loop
  • Compliance violation in motion

→ Replace immediately, then ask why it emerged

From Feral IT to Feral AI

We've been studying this emergence pattern long before it had an AI label. The shape is familiar. The speed is new.

🗃️ Feral IT (then)

  • The artifact: Spreadsheets, Access databases, personal scripts, departmental SaaS.
  • The builder: A power user who knew just enough.
  • The spread: Email attachments and "ask Bob for a copy."
  • The blast radius: Bounded by the team and the tool.
  • The risk: Stale data, lost backups, key-person dependency.
  • The detection: Network logs, SaaS inventory, expense reports.

🐺 Feral AI (now)

  • The artifact: Custom GPTs, agents, prompt libraries, Copilot-generated scripts.
  • The builder: Anyone with a chat window and a problem.
  • The spread: A link, a share, a "try this prompt" in Slack.
  • The blast radius: Any data the model has ever seen, and any system the agent can act on.
  • The risk: Data absorbed into model weights, hallucinated outputs, autonomous actions, supply chain in dependencies.
  • The detection: Much harder — it lives inside sanctioned apps and looks like normal user activity.

Feral AI is the same emergence engine — workers solving real problems faster than formal IT can — running on a more powerful substrate. The methodology that worked for Feral IT (discover, evaluate, channel, scale) works here. The execution has to be faster and the choice architecture has to be sharper.

Our Methodology

Behavioral science is the only governance approach that survives contact with Feral AI. Restrictions get routed around in an afternoon. Choice architecture compounds value and focus.

🧠 Nudge Theory for AI Adoption

We apply the same behavioral economics that won the Nobel Prize to the question every CIO is now living: how do you get an entire workforce to choose the sanctioned AI path on their own?

Channel, don't ban. When approved AI tools are offered with the right defaults, unauthorized use drops by an estimated 89%. The lever is design, not policy.

We help organizations design defaults, social proof, and friction-reduction strategies that make the sanctioned platform the easy answer — and the unsanctioned shortcut the harder one.

🏛️ Choice Architecture for Feral AI

Most governance programs respond to Feral AI by trying to eliminate it. We design choice architectures where inclusion is the path of least resistance.

Default Integration Pathways: Approved tools come pre-wired to corporate data sources. Amnesty programs convert hidden tools into catalogued assets. "Innovation pioneer" recognition replaces "rule breaker" framing.

The result is a discovery, evaluation, and integration loop that runs continuously — at the speed Feral AI actually moves — instead of an annual audit that's already out of date when it ships.

🤖 Generative AI as the Lens

We use AI to study AI. Pattern detection across prompts, agents, and shadow workflows surfaces the emergent systems so leaders can see what's actually being built — and predict where the next workaround will appear.

Telemetry first. You can't channel what you can't see. We help build the signal layer that turns Feral AI from rumor into roadmap.

💡 Rewarding Innovation

Your employees are solving real problems with their use of AI. Filling these gaps and removing points of friction can unlock trapped potential.

We help you build a self-reinforcing virtuous cycle where innovation in areas where tech has not previously worked is recognized and rewarded, which drives creativity and productivity to solve new business problems.
What previously would require a software engineer weeks to build can now done in minutes by a manager, clerk, or sales person. Good ideas are being unlocked and you can help them scale.

Advisory for CIOs

Engagements designed for IT leadership teams confronting a Feral AI they didn't sanction and can't yet see.

🔍 Feral AI Discovery

A structured inventory of the AI tools, agents, custom GPTs, prompts, and AI-glued workflows already running in your business. Output: a map of what exists, what it touches, and where the risk concentration sits.

🏛️ Governance Choice Architecture

Design the sanctioned path so it wins. We help your team build defaults, approval pathways, and reward structures that make the right AI choice the easy AI choice.

📈 Adoption Programs

Nudge-based rollout for approved AI platforms. Pilot design, communications, social proof loops, and measurement — so sanctioned adoption compounds instead of stalling at 15%.

🎓 Executive Workshops

Half-day and full-day workshops for CIO leadership teams and their direct reports. Align on the Feral AI thesis, walk the lifecycle, and leave with a 90-day plan you can actually run.

📊 Measurement & Telemetry

Track sanctioned vs. unsanctioned use over time. We help you instrument the signal layer that proves the 89% effect inside your own org — and surfaces the next workaround before it scales.

🔄 Integration & Scale-Up

For the Feral AI tools worth keeping, we help your platform teams formalize them — bringing logging, identity, and lifecycle into systems your CISO and your auditors can sign off on.

Where most engagements start: a 4–6 week Discovery + Choice Architecture sprint, scoped to one business unit. Quick enough to prove the model, deep enough to find the things that scare you.

Speaking & Executive Briefings

Keynotes, executive briefings, and workshops on Feral AI, behavioral security, and the governance of emergent systems — for CIO summits, security conferences, and leadership offsites.

🎤 AWS re:Inforce 2025

Transforming Security Compliance with Nudge Theory

Traditional security approaches often create friction between security teams and builders, increasing costs while reducing innovation. This presentation explores how nudge theory — a Nobel Prize-winning behavioral science framework — can transform security compliance at scale.

"Learn how subtle changes in choice architecture, default settings, and communication methods dramatically improve security outcomes while reducing engineering burden."

Through real-world case studies, discover how organizations achieve higher compliance rates and faster remediation without additional headcount, while fostering a positive security culture — the same mechanism that, applied to AI adoption, drives the 89% shift toward sanctioned tools.

🎯 Topics for CIO Audiences

Feral AI: The Emergent Agents You Didn't Sanction. What's actually being built inside your business right now, why it's growing 10× faster than Shadow IT ever did, and how to channel it.

Behavioral Security at Scale. Nudge theory applied to compliance, AI adoption, and the chronic friction between security and engineering.

Defending in the Mythos Era. What AI-accelerated vulnerability discovery means for patch programs, attack surface, and the value of governed code.

Drawing on 20+ years protecting critical systems at Amazon, HPE, and DXC Technology, including experience with state-sponsored threats and major security incidents.

🎯 Booking & Briefings

Available for keynotes, panels, executive briefings, and CIO leadership team workshops on Feral AI, behavioral security, and the governance of emergent systems.

Contact us to discuss speaking or a private briefing for your leadership team.

About

Feral Systems is led by Logan Browne, who has spent two decades at the intersection of security, behavioral science, and the systems people actually build to get work done.

Logan Browne is Senior Manager for Amazon Global Media and Entertainment Security, specializing in the intersection of behavioral science, information systems, and organizational transformation. With over 20 years of experience in cybersecurity, technical leadership, and protecting critical systems from sophisticated adversaries, he brings a unique perspective to understanding how emergent systems develop in complex organizations — and how that pattern is now playing out, at unprecedented speed, in the form of Feral AI.

His work combines expertise in cyber security management, threat management fusion, security operations, and technical team leadership with practical experience building and scaling security engineering teams at Amazon, DXC Technology, and Hewlett Packard Enterprise. He holds an MBA from UC Davis with a concentration in Management of Organizations, a Bachelor's in Computer Science from Hiram College (Cum Laude with Departmental Honors), and is a Certified Information Systems Security Professional (CISSP).

Through research, advisory work, and speaking, Logan helps CIOs and IT leadership teams understand why informal AI systems emerge and how to leverage behavioral science to optimize rather than eliminate these valuable adaptations. His extensive background includes building multiple "two pizza teams" at Amazon, leading global incident response during major security events, and developing innovative security solutions that have protected intellectual property and systems for some of the world's largest technology companies.

Ready to Channel Your Feral AI?

The systems are already there. The question is whether you see it, channel it, and scale the winners — or wait for the first incident to make the introduction.

📧 Advisory Inquiries

For Discovery sprints, governance design, and executive workshops with CIO leadership teams.

🎤 Speaking & Briefings

For keynotes, panels, and private executive briefings on Feral AI, behavioral security, and the governance of emergent systems.

Schedule an Advisory Briefing