What Is the CIA Triad in Cyber Security? Confidentiality, Integrity and Availability Explained

The CIA triad in cyber security stands for confidentiality, integrity, and availability, the three pillars that guide how organizations protect information. Every major breach you’ve heard of failed at least one of these three. Understanding which one helps you see exactly what went wrong, and what to fix.

What is the CIA triad in cyber security?

The CIA triad is confidentiality, integrity, and availability, three properties that together define what it means for information to be secure. Confidentiality means only authorized people can see data. Integrity means data stays accurate and unaltered. Availability means authorized users can access it when needed.

These three pillars work as a checklist for evaluating any security control or incident. A stolen password threatens confidentiality. A tampered database record threatens integrity. A ransomware attack that locks systems threatens availability. Most real incidents touch more than one pillar at once, which is exactly why the model stays useful for classifying and prioritizing risk.

Where did this come from? A look at the origin

The CIA triad doesn’t have one clean origin story. Confidentiality was formalized in a 1976 US Air Force study, integrity was explored in a 1987 paper by David Clark and David Wilson, and availability became widely discussed around 1988, with the abbreviation itself credited to Steve Lipner around 1986.

The concept traces back further still to the 1972 Anderson Report, written for the Air Force’s Electronic Systems Division, which first raised computer security as a distinct concern. Confidentiality mattered first because early computing was defense-focused. Integrity became a commercial priority once banks and financial services needed assurance that electronic records weren’t silently altered. Availability entered the conversation later, once early denial-of-service incidents in the late 1980s showed that a system being inaccessible was its own category of failure. By the late 1990s, security professionals had unified all three into the single “CIA triad” name used today. Most content skips this piecemeal history in favor of a tidy single date, but the honest version matters: it shows the model wasn’t designed top-down, it emerged from three separate real-world pressures converging over two decades.

Confidentiality, integrity, and availability: real failures, one for each

Three real, well-documented incidents show each pillar failing in isolation. Equifax lost confidentiality in 2017. A 2013 AP Twitter hack broke integrity. WannaCry broke NHS availability in 2017.

Pillar	Incident	What Happened
Confidentiality	Equifax, 2017	Unpatched Apache Struts flaw exposed 147.9 million records
Integrity	AP Twitter hack, 2013	False tweet about a White House explosion dropped the Dow 147 points in minutes
Availability	WannaCry, 2017	Ransomware hit 81 NHS trusts, cancelling ~19,000 appointments

The Equifax breach happened because a known vulnerability, CVE-2017-5638, sat unpatched for two months, letting attackers extract Social Security numbers, birth dates, and addresses for 147.9 million Americans, plus 15.2 million UK citizens. That’s a pure confidentiality failure: data got seen by people who shouldn’t have seen it, nothing was altered or made unavailable. The 2013 case is different in kind entirely. The Syrian Electronic Army compromised the Associated Press’s Twitter account and posted a fabricated claim that President Obama had been injured in explosions at the White House. No data was stolen and nothing went offline, but the market briefly treated false information as true, and the Dow dropped 147 points within minutes before recovering. That’s integrity failing on its own: the information looked legitimate but wasn’t. WannaCry shows the third pillar cleanly. Ransomware locked systems across 81 NHS trusts and 595 GP practices in May 2017, forcing roughly 19,000 appointment cancellations and costing the NHS £92 million, according to the Department of Health and Social Care’s own report. Crucially, NHS Digital confirmed no patient data was stolen or altered, this was availability failing in isolation, systems simply became unusable. Seeing all three side by side, rather than one generic combined example, makes the distinction between the pillars far easier to actually apply during incident classification.

Bringing it together: how encryption, hashing, and backups each defend one pillar

Each core security control maps cleanly to one pillar. Encryption protects confidentiality by making data unreadable without a key. Hashing protects integrity by detecting any change to data. Backups protect availability by ensuring data survives even if the primary system fails.

Encryption doesn’t stop someone from stealing a file, it stops them from reading it once stolen. Hashing works differently: a hash function generates a fixed fingerprint of data, and any tampering changes that fingerprint, which is how software downloads and digital signatures verify nothing was altered in transit. Backups solve a completely separate problem, they guarantee that even total system loss doesn’t mean permanent data loss. Understanding which control maps to which pillar helps you spot gaps: a business with strong encryption but no tested backup strategy has covered confidentiality while leaving availability exposed.

The pillars can conflict with each other, here’s how

Strengthening one pillar can weaken another. Enabling broad remote access improves availability but increases confidentiality risk if access controls aren’t tightened at the same time. Heavy encryption protects confidentiality but can slow legitimate access, hurting availability.

A concrete example: a hospital granting emergency-room staff instant access to any patient record improves availability during a crisis, but it widens the pool of people who can see sensitive data, weakening confidentiality. The fix isn’t picking one pillar over another, it’s designing controls that manage the trade-off deliberately, like time-limited emergency access that logs and expires automatically. Recognizing that these three properties pull against each other, rather than treating them as three independent boxes to tick, is what separates a security program that works from one that just looks complete on paper.

Do different industries prioritise the three pillars differently?

Yes. Government and intelligence sectors typically prioritize confidentiality. Financial institutions prioritize integrity, since a single misplaced decimal can be catastrophic. Healthcare, e-commerce, and critical infrastructure prioritize availability, where downtime carries safety or revenue consequences.

This isn’t arbitrary, it follows directly from what a sector’s regulatory or operational failure actually looks like. Healthcare must comply with HIPAA in the US, which drives confidentiality investment, but a hospital’s clinical systems being unavailable during an emergency is arguably the more immediate patient-safety risk, which is why healthcare often ends up balancing both pillars heavily rather than picking one. Financial services regulators care intensely about transaction accuracy, so integrity controls like reconciliation and audit trails get disproportionate investment there. None of the three pillars is ever fully ignored in any sector, but knowing which one your industry weighs heaviest helps you allocate a limited security budget where it actually reduces the most real-world risk.

Does the CIA triad still hold up against AI and prompt injection?

Yes, the three categories still cover every documented AI attack pattern. Confidentiality breaks when prompt injection extracts system prompts or chat history. Integrity breaks when injected content hijacks a model’s output or tool calls. Availability breaks when attackers flood a model with expensive prompts to exhaust resources.

Prompt injection was first identified as a distinct threat class in 2022, and researcher Johann Rehberger’s widely cited paper “Trust No AI: Prompt Injection Along The CIA Security Triad” catalogs real, documented exploits across production systems from OpenAI, Microsoft, Anthropic, and Google, mapped cleanly onto the three pillars. A real proof-of-concept against Google Gemini in Google Docs used a hidden prompt injection payload to render a hyperlink connecting the victim directly to an attacker via Google Meet, a confidentiality and integrity failure combined. Microsoft 365 Copilot in Outlook has faced similar conditional injection attacks exploiting its access to organizational structure data. Prompt injection is now ranked the top threat in the OWASP Top 10 for LLM Applications. Some researchers propose extending the model with Trust and Autonomy for AI-specific cognitive risks, but the original three categories still draw clean, useful lines around every attack type documented so far, meaning businesses adopting AI tools can and should keep using CIA-triad thinking to evaluate those risks rather than treating AI security as an entirely separate discipline.

Is this a complete model? The Parkerian Hexad

No, the CIA triad has known gaps, which is why Donn B. Parker proposed the Parkerian Hexad in 1998, adding possession or control, authenticity, and utility as three additional, non-overlapping properties information can lose independently of confidentiality, integrity, or availability.

Utility is the clearest example of what the triad misses. An encrypted backup can be fully confidential, unaltered, and technically available, yet still useless if the decryption key is lost, that’s a utility failure the CIA triad has no vocabulary for. Possession or control covers a related gap: ransomware that encrypts your own files without ever copying them doesn’t breach confidentiality in the classic sense, but you’ve lost control over your own data. The Hexad isn’t a replacement for the triad, it’s a more granular lens for post-incident forensics and complex scenarios, particularly useful when investigating exactly which specific property an attack targeted.

How does this connect to your UK GDPR obligations?

GDPR Article 32 explicitly names confidentiality, integrity, availability, and resilience as required security properties, meaning the CIA triad isn’t just a security convention, it’s written directly into UK and EU data protection law as a compliance requirement.

Article 32(1)(b) requires controllers and processors to ensure “the ongoing confidentiality, integrity, availability and resilience of processing systems and services” for any personal data they handle. This means a UK business can map its existing security controls directly onto its GDPR compliance evidence: encryption demonstrates confidentiality measures, integrity checks and audit logs demonstrate integrity measures, and tested backup and recovery capability demonstrates both availability and the added resilience requirement. If you’re already organizing your security program around the CIA triad, you’re most of the way toward documenting Article 32 compliance, since the language regulators use and the language security teams already speak are, in this case, nearly identical.

A practical way to apply this if you’re not running an enterprise security team

Run every security decision through three quick questions: does this protect confidentiality, integrity, or availability, and which one matters most for this specific system? Small businesses without a dedicated security function can use this as a lightweight prioritization filter without needing a formal framework first.

Start by listing your critical systems, customer records, financial data, email, and tag each with its most important pillar based on what would hurt most if it failed. A customer database probably weights confidentiality highest. Your invoicing system probably weights integrity highest. Your website probably weights availability highest. This single exercise, done honestly, tells you where to spend limited security budget first. Cyber Security Solutions Ltd uses exactly this triage approach with smaller clients who don’t have an in-house CISO, and it consistently surfaces the one or two systems that actually need attention first, rather than spreading effort evenly across everything.

Conclusion

The CIA triad still does what it was built to do: give you a fast, honest way to classify what actually broke, whether that’s a stolen database, a manipulated record, or a locked-up system, and it holds up even against AI-era threats like prompt injection. Use it to triage your own systems, starting with whichever pillar matters most for your highest-risk data. If you want help mapping your security controls against confidentiality, integrity, and availability, Cyber Security Solutions Ltd can walk through it with you at cybersecuritysolutionsltd.com.