Is Your Business Ready for AI-Powered Attacks? A Practical Checklist
AI attack readiness is a plain assessment of whether you would notice a change to your internet-facing systems within a day rather than a quarter. It is not a new toolset: know every exposed asset, monitor for new subdomains and open ports daily, detect configuration drift in hours, and keep certificates and software current.
AI-enabled adversary operations increased 89% year-over-year according to CrowdStrike's 2026 Global Threat Report. An autonomous AI agent reached #1 on HackerOne's US leaderboard in 2025, and an open-source offensive framework was used in a campaign against internet-facing appliances in 55 countries. None of this changes what good security looks like — it changes how much the cadence matters. This guide is a practical 10-point checklist for assessing where you stand.
- Current AI threat landscape: 89% increase in AI-enabled operations
- XBOW: the autonomous agent that topped HackerOne's leaderboard
- CyberStrikeAI: confirmed attacks across 55 countries
- What the CSA's "Mythos-ready" guidance actually asks for
- A 10-point AI Attack Readiness Checklist with actionable guidance
- How DriftAlarm addresses each checklist item
The AI Threat Landscape in 2026
The CrowdStrike 2026 Global Threat Report documented a fundamental shift in the cyber threat landscape. AI-enabled adversary operations increased 89% year-over-year, driven by three converging factors: the availability of powerful AI models, the proliferation of open-source offensive AI tools, and the dramatic reduction in skill required to launch sophisticated attacks.
An 89% annual increase means AI-enabled adversary activity nearly doubled in a year. We are deliberately not projecting that rate forward — a single year's growth is not a trend line, and the number that matters for planning is not next year's volume but how quickly you would notice a change to your own perimeter.
The nature of attacks is changing as fundamentally as their volume. AI-powered attacks are characterized by speed (exploitation within minutes of discovery), breadth (thousands of targets scanned simultaneously), and sophistication (multi-step attack chains that would previously require skilled human operators). The traditional assumption that small businesses are not worth targeting has been invalidated by the near-zero marginal cost of AI-powered scanning.
XBOW: AI Pentesting Is Production-Ready
In June 2025, an autonomous AI agent called XBOW reached #1 on HackerOne's US bug bounty leaderboard. This was not a controlled demo. XBOW competed against human security researchers on real bug bounty programs and reached the top of the US ranking.
XBOW demonstrated that AI pentesting has crossed the threshold from research project to production-ready capability. The agent autonomously:
- Identified target applications and their technology stacks
- Performed reconnaissance and attack surface enumeration
- Discovered and validated exploitable vulnerabilities
- Generated proof-of-concept exploits
- Submitted properly formatted bug reports
XBOW's success on HackerOne demonstrates a defensive capability — finding bugs before attackers do. But the same technology, applied offensively, creates an autonomous attack agent that can discover and exploit vulnerabilities without human guidance. The techniques are identical. Only the intent differs.
XBOW's achievement signals that the era of human-only penetration testing is ending. AI agents can now match or exceed human researchers in vulnerability discovery for many common vulnerability classes. For defenders, this means the volume and speed of probing against your attack surface is about to increase dramatically.
CyberStrikeAI: An Open-Source Framework Used in a Real Campaign
Where XBOW is the defensive application, CyberStrikeAI shows the offensive one. It is an open-source offensive framework published on GitHub — not a bespoke nation-state tool. Research from Team Cymru, reported in March 2026 with initial findings from Amazon Threat Intelligence, documented a suspected Russian-speaking actor using it to compromise more than 600 Fortinet FortiGate appliances across 55 countries between 20 January and 26 February 2026.
Two details matter for planning. First, the 55 countries describe the victim spread of that one campaign, not the tool's installed base — a distinction worth keeping, because the campaign figure is the one that is actually evidenced. Second, the entry point was a class of internet-facing appliance, which is exactly the kind of asset that tends to sit outside a small team's patching routine.
The notable thing about this campaign is not that the tooling was advanced. It is that the framework was publicly available, and the way in was a reachable, unpatched edge appliance. That is a coverage problem before it is a sophistication problem — which is good news, because coverage is the part you can actually fix.
Building a "Mythos-Ready" Security Program
In April 2026 the Cloud Security Alliance published The "AI Vulnerability Storm": Building a "Mythos-ready" Security Program, setting out roughly eleven priority actions organized by horizon — this week, the next 45 days, and the next 12 months. It is worth reading in full; the link is in the sources below.
Three of its themes carry most of the weight for a small IT team, and they are the ones this guide is built around. The framing below is ours, not CSA's:
Your external attack surface is being scanned by AI-powered tools continuously. Design your security program around the assumption that every exposed service, port, and configuration will be tested for weaknesses daily — not annually.
CrowdStrike put average eCrime breakout time at 29 minutes in its 2026 report, with the fastest it has ever observed at 27 seconds. Treat 29 minutes as the planning number and the 27 seconds as the tail. Either way the implication is the same: shift investment from periodic assessment to continuous monitoring, and aim to detect attack-surface changes within hours.
Every internet-facing service is a target. Reduce your external footprint to the minimum required for business operations. Remove end-of-life software, close unnecessary ports, clean up unused DNS records, and decommission forgotten services. The best defense against AI-powered scanning is having fewer targets to scan.
Assess Your AI Readiness Now
Use the checklist below to evaluate your organization. Then see how DriftAlarm addresses each item with continuous, automated monitoring of your external attack surface.
10-Point AI Attack Readiness Checklist
Score yourself honestly. Each item is binary — you either have the capability or you do not. A "sort of" or "we are working on it" counts as a no. AI-powered attackers will not wait for your roadmap.
Can you enumerate every domain, subdomain, IP address, and cloud resource your organization exposes to the internet? This includes shadow IT, forgotten staging environments, marketing microsites, and legacy systems. If you cannot list them all, AI scanners already know more about your surface than you do.
New subdomains can appear from dev teams spinning up test environments, marketing launching campaign sites, or attackers compromising your DNS. Daily monitoring ensures new assets are discovered and assessed within 24 hours — not at the next quarterly review.
When a port opens, a TLS certificate downgrades, or a DNS record changes, do you know about it the same day? Configuration drift creates the conditions that AI-powered tools exploit. Detecting drift within hours means you can remediate before automated scanners weaponize the change.
Annual pentests provide depth but miss 364 days of changes. Automated vulnerability scanning provides continuous coverage that catches new CVEs as they are published — not months later. With mean time-to-exploit now reported as negative — Mandiant's M-Trends 2026 puts it at roughly -7 days, meaning exploitation frequently precedes public disclosure — daily scanning is the minimum viable cadence.
Dangling CNAME records — DNS entries pointing to decommissioned cloud resources — enable subdomain takeover attacks. An attacker can claim the orphaned resource and serve malicious content under your domain name. AI-powered tools actively scan for dangling CNAMEs at scale.
Expired certificates, weak cipher suites, and deprecated protocol versions are all signals that AI scanners use to identify poorly maintained targets. Certificate issues also erode user trust and can break functionality. Monitor expiration dates, protocol versions, and certificate chain validity continuously.
End-of-life software receives no security patches. It is permanently vulnerable to any exploit discovered after its end-of-life date. AI vulnerability discovery accelerates the rate at which new exploits are found. Running EOL software on the internet is equivalent to leaving a door permanently unlocked.
Ports opened for maintenance, debugging, or deployment frequently remain open indefinitely. Each open port is an additional entry point that AI scanners will probe. Alerting on new open ports ensures that temporary exceptions do not become permanent attack surface.
When your CEO or board asks about your security posture, can you provide a data-driven answer? Risk scores, vulnerability counts, drift trends, and patch coverage metrics tell a clear story. "I think we are fine" is not a security posture assessment — it is a hope.
With AI generating more vulnerability findings than ever, human-only triage creates a bottleneck. AI-powered risk scoring combines CVSS severity, EPSS exploitation probability, asset criticality, and environmental context to prioritize what matters most — not just what scored highest on a generic scale.
How DriftAlarm Addresses Each Checklist Item
DriftAlarm was built for the threat environment described in this guide. Here is how the platform maps to each item in the AI Attack Readiness Checklist:
| # | Checklist Item | DriftAlarm Capability |
|---|---|---|
| 1 | Know your full attack surface | Automated discovery scanning enumerates subdomains, IPs, ports, services, and technologies |
| 2 | Monitor for new subdomains daily | Daily DNS enumeration with alerts for new subdomain discovery |
| 3 | Detect configuration drift within hours | 32 drift detection rules compare every scan against automated baselines |
| 4 | Automated vulnerability scanning | Daily vulnerability scanning across your internet-facing assets |
| 5 | Clean DNS records | Dangling CNAME detection and DNS drift monitoring across all records |
| 6 | SSL/TLS certificates current | Certificate expiration tracking, protocol version monitoring, cipher suite analysis |
| 7 | EOL software removed | Technology fingerprinting with end-of-life detection and alerts |
| 8 | Alerts for new open ports | Port drift detection alerts on any new open port with service identification |
| 9 | Data-driven security posture | Risk scores, trend charts, vulnerability counts, and weekly security reports |
| 10 | AI-powered prioritization | AI risk analysis combining CVSS, EPSS, asset context, and environmental factors |
Frequently Asked Questions
How is preparing for AI-powered attacks different from ordinary security hygiene?
Mostly it isn't — and that is the useful finding. Every item on a readiness checklist is something security teams have recommended for a decade: know your assets, watch for changes, patch what faces the internet, retire software that no longer gets updates. What changed is the cost of skipping them. Automation has made the gap between "we check quarterly" and "we check daily" matter more than it used to, so cadence is where readiness is actually won or lost.
We have no security staff — where do we start on this checklist?
Start with item one and do not move on until it is done. A complete list of what your company exposes to the internet — every domain, subdomain, public IP, and cloud resource, including the ones inherited from projects that ended — makes the other nine items possible and several of them unnecessary. Most small IT shops discover assets they had forgotten at this step, and removing them is faster than defending them.
Is an annual penetration test still worth it?
Yes, for depth — but not as your only assessment. A penetration test brings human judgment to business logic, chained attack paths, and the things automation reads past; that value has not gone away. What it cannot do is cover the 364 days between tests, and most of what appears on your perimeter appears in that window. Run the test for depth and a continuous check for coverage.
How quickly do we actually need to detect a change to our attack surface?
Within a day is the working standard for anything internet-facing. A new subdomain, a port left open after maintenance, or a certificate that quietly weakened is discoverable by anyone scanning the internet within minutes of the change, so the number that matters is how long it stays unknown to you — not how thorough your last review was. Large enterprises bought that cadence years ago; DriftAlarm's premise is that a smaller company can have it too, at a flat monthly price.
Sources
- CrowdStrike, 2026 Global Threat Report (February 2026) — the 89% year-over-year increase in AI-enabled adversary activity, and the 29-minute average / 27-second fastest eCrime breakout times.
- XBOW, Top 1: how XBOW did it (June 2025) — reaching #1 on HackerOne's US leaderboard.
- Team Cymru research with initial findings from Amazon Threat Intelligence, reported March 2026 — CyberStrikeAI used against 600+ FortiGate appliances across 55 countries, 20 Jan–26 Feb 2026.
- Cloud Security Alliance, The "AI Vulnerability Storm": Building a "Mythos-ready" Security Program (April 2026).
- Mandiant / Google Cloud, M-Trends 2026 — mean time-to-exploit reported as negative.
Start Assessing Your AI Readiness Today
Every item on this checklist that you cannot answer "yes" to is a gap that AI-powered attackers will find. DriftAlarm gives you the continuous visibility needed to close those gaps — starting with a free 30-day trial of the full platform.