A web application firewall sits between the internet and your application. It inspects HTTP traffic, applies managed and custom rules, and decides whether to allow, challenge or block each request. When it is configured well, it absorbs a large share of automated attack traffic before it reaches your code. When it is configured badly, it creates two problems at once: it gives the business a false sense of protection and it breaks legitimate customer journeys.
Many WAF failures are not engine failures. They are configuration failures that accumulate quietly after launch. This guide sets out the risks worth ranking, the controls that keep a configuration honest, and the questions a board or management team should ask before signing off on a WAF project.
What a WAF can and cannot do
A WAF is a control at the edge of the application. It applies signature and anomaly rules, rate-limits abusive clients, challenges suspicious sessions and logs what it sees. It cannot fix a broken authorisation check, patch a vulnerable library, remove a hard-coded secret or stop an attacker who already has valid credentials. Treating the WAF as a substitute for secure development is a common and expensive misconception.
The risk categories
1. Coverage that is not real
The WAF only protects traffic that actually passes through it. In practice, teams discover direct-to-origin access, an old staging hostname, a mobile API subdomain, a partner integration that calls the server by IP address, or a websocket endpoint excluded from inspection. An attacker who can reach the origin directly can ignore the WAF entirely.
Evidence to check: resolve the origin IP and send a request with the correct host header; review DNS history; list every hostname that serves production data; confirm the origin firewall accepts traffic only from the WAF provider.
2. Over-blocking legitimate traffic
Managed rules switched straight to block mode are the classic cause. Checkout submissions, file uploads, rich-text editors, search queries with punctuation, and mobile app payloads can all look suspicious to a generic rule set. The result is a spike in 403 responses, failed payments and support tickets that nobody connects to the WAF release.
Over-blocking is visible to customers, happens during peak trading, and erodes trust in future security changes.
3. Under-blocking and rule drift
The opposite failure is a configuration that has been weakened until it passes everything. Exclusions added during a launch are rarely removed. A rule disabled during an incident is often never re-enabled. New application features, such as a new upload endpoint or a GraphQL API, may not be covered by the existing rule set at all.
Anomaly thresholds also drift. A team lowers the threshold to stop false positives, then forgets that the same threshold now permits common attack patterns.
4. Logging and alerting gaps
If rule hits are not logged with a request identifier, the team cannot correlate a block with an application error or a customer complaint. If logs are kept for only a few days, incident review becomes guesswork. If nobody is alerted when a new rule starts blocking a large share of traffic, the first alert comes from the contact centre.
5. No named owner and no change control
WAF configuration is a living system. Someone must approve rule changes, review exclusions and own the relationship with the provider. Without that ownership, every urgent fix becomes a permanent exception, and the rule set slowly becomes a record of past incidents rather than a control.
6. TLS and origin exposure
Many teams terminate TLS at the WAF but leave the origin accepting plain HTTP, or leave an admin panel reachable on the public internet. The WAF then protects the shopfront while the back door stays open.
7. Dependency on the rule provider
Managed rule sets are software supply chain. Updates change behaviour, providers have outages, and rule syntax is not portable. A provider outage can take the site down if the WAF fails closed without a tested fallback. A rule update can break a checkout path that worked yesterday.
8. Treating bot management as a WAF feature
Credential stuffing, scraping and fake account creation need rate limits, bot scoring, device signals and multi-factor authentication. A signature WAF alone will not stop a slow, distributed login attack that stays under request thresholds.
Judging likelihood and impact
Use a three-level scale. For likelihood, ask whether the issue is already present and whether an outsider could reach it without insider knowledge. For impact, separate revenue, customer data, availability and regulatory exposure.
- Direct origin bypass: medium likelihood, high impact.
- Over-blocking at checkout: medium likelihood, high impact because it hits revenue during peak hours.
- Rule drift and stale exclusions: high likelihood, medium impact. It accumulates in every configuration that lacks review.
- Logging gaps: high likelihood, medium impact. The real cost appears during incident response.
- Provider outage with fail-closed behaviour: low likelihood, high impact. It is an availability risk, not just a security risk.
Controls that hold up
- Start in log mode. Run new managed rules in detection mode long enough to see what they would block, then move to block mode route by route.
- Lock the origin. Allow only the WAF provider's egress ranges, enforce TLS, and remove public admin access.
- Keep an exclusion register. Every exclusion needs an owner, a reason, a ticket and an expiry date.
- Test from the outside. Synthetic checkout, login and search journeys should run through the WAF after every rule change.
- Alert on behaviour, not just attacks. Watch the 403 rate by route, rule ID and user agent.
- Review quarterly. Re-read rules, remove dead exclusions, confirm coverage of new endpoints and re-test the origin lock.
- Write a runbook. Include how to switch a rule to log mode, who can approve it, and how to communicate a false-positive outage.
- Test the fallback. Decide explicitly whether the WAF fails open or closed, and rehearse the provider-outage path.
What to measure
- Share of production hostnames routed through the WAF.
- Blocked and challenged requests by rule ID and route.
- False positives confirmed by support, with time to resolution.
- Age of the oldest open exclusion.
Board questions
- Which production hostnames are not behind the WAF, and why?
- Who owns the rule set, and when was it last reviewed?
- How do we know a rule change has not broken checkout?
- Can we reconstruct a blocked request from logs six months later?
- What happens to the site if the WAF provider has an outage?
- Which attacks are we expecting the WAF to stop that it cannot stop?
Where this fits in our work
If you need the configuration reviewed rather than replaced, start with Managed Technology and Cybersecurity. Our pricing page explains how scoping works. For the recovery side of the same problem, read the malware cleanup prevention checklist.
Sources
- OWASP ModSecurity Core Rule Set, accessed 17 November 2026.
- NIST SP 800-41 Rev. 1, Guidelines on Firewalls and Firewall Policy, accessed 17 November 2026.
- Cloudflare, Web Application Firewall documentation, accessed 17 November 2026.
- AWS, What is AWS WAF?, accessed 17 November 2026.
- OWASP Top 10, accessed 17 November 2026.
Responses (0)
No comments yet. Add the first one below.
Leave a response
Your email address will not be published. Required fields are marked *.