A Decade of Cybersecurity Jobs: What's Changing and What Comes Next?

Loading the ten-year hiring overview…

Summary

Prioritise cloud identity and data protection, application security tied to software delivery, and automation for vulnerability management and incident response. In governance, risk and compliance (GRC), prioritise technical control testing and evidence collection. AI Security Engineer, AI Security Specialist and agentic security roles now explicitly ask engineers to secure model integrations, agent permissions and AI development workflows. Roles limited to routine alert triage, scanner reporting or evidence administration are more exposed to automation than roles that investigate incidents, test controls or engineer fixes.

Those recommendations combine current hiring-manager priorities, recent UK job-ad trends and documented changes in security tooling. The work includes configuring permissions, testing application access controls, building security checks into delivery pipelines, joining vulnerability findings to actual assets, and validating AI-generated queries or fixes. The sections below show how those tasks differ from older job descriptions and how to use AI in specific workflows.

Main findings

• AI security is appearing as a dedicated engineering specialism. The reviewed adverts ask for agent permissions, secure retrieval, adversarial testing, runtime detection and AI governance. Other roles use AI to find vulnerabilities but still require code review, exploitability assessment and remediation.
• Routine Tier 1 SOC jobs are exposed to staffing cuts. Alert enrichment, initial classification and ticket drafting are already automation targets. Develop investigation, detection tuning and response skills beyond processing the queue.
• Cloud, code and automation were already present a decade ago. The 2026 examples specify delivery-pipeline controls and AI assistance.
• Cloud security and security engineering have direct hiring support. Prioritise identity permissions, secure configuration, application access controls and security tests in continuous integration and delivery (CI/CD).
• Automation needs programming and security data skills. Learn Python, SQL or your team's query language, APIs, structured scanner output and asset inventories. These connect findings to affected systems, owners and remediation.
• AI is being used to investigate findings, draft fixes and write queries. The required security work includes reproducing the finding, testing the patch and checking the query against the actual logs.
• Risk and assurance remain relevant. Learn to test a control against current system evidence and track exceptions, including the permissions and data flows of AI agents.

What the job looked like in 2016–2017, and what appears in 2026

Earlier roles already combined technical work and ownership

A November 2016 VSR advert sought penetration testers for application and network assessments, red-team exercises and social engineering. The advert also requested collaboration.

Square's December 2016 product-security advert asked for ownership of product security and decisions about mitigating risk while working with product teams. Blink Health's January 2017 advert already combined AWS infrastructure, Linux, application security, policy/governance and coding or automation skills.

These examples establish that cloud, software security and automation were not new additions in 2026. They also show why a ten-year comparison cannot describe every earlier role as network administration or manual scanning.

Current adverts describe delivery controls and AI-assisted operations

Elucid's August 2026 senior-security advert specifies threat modelling, code and design review, security testing in CI/CD, AWS hardening and audit work. A single role spans the application, its deployment and the evidence needed to assess its controls.

Odyssey's June 2026 advert explicitly calls for using AI in threat detection, vulnerability analysis and incident response. That is an expectation about how the security engineer performs the work, not merely an employer describing an AI product.

The examples illustrate individual jobs; the chart measures wording across the larger sample. In the pooled early and recent windows, AI/ML language and senior, staff or principal wording became more common. They do not establish that every team adopted the same operating model.

The emerging AI security engineer role: what employers ask for

Dedicated AI security jobs now appear under titles including AI Security Engineer, AI Security Specialist, Agentic AI Security Specialist and AI Security Architect. The seven selected descriptions reviewed here show distinct responsibilities: securing AI systems, testing agents, governing adoption, and using AI in existing security work. They establish that these jobs exist, not their market share or growth rate. These examples are separate from the HN chart dataset.

Building security into AI platforms and agents

The Esko platform-security description asks engineers to build tenant-, user-, document- and tool-level access controls, secure retrieval, isolate execution and record security activity. It requests production software engineering in Python, TypeScript or a similar language. The role is building reusable controls for product teams.

Deel's AI Security Specialist advert combines AI adoption reviews with securing Model Context Protocol (MCP) servers, data flows, connectors and AI-assisted development. HCLTech's Agentic AI Security Specialist advert adds Python/Go security components, detection of agent misuse, and mechanisms to isolate or stop unsafe execution.

Skills to prioritise: software engineering, identity and authorisation, API security, tenant isolation, permission-aware retrieval and runtime monitoring. A relevant implementation is an assistant whose search tool enforces the requesting user's document permissions and whose action tool cannot execute outside its assigned scope. Test both allowed operations and attempts to cross those boundaries.

Red teaming and evaluating AI behaviour

ClearRoute's AI Security Engineer (Agentic Systems) advert requests direct and indirect prompt-injection testing, tool-call manipulation, memory/context poisoning and repeatable evaluations of outputs that vary between runs. It also includes incident procedures, governance and training other engineers.

Skills to prioritise: threat modelling, adversarial testing, evaluation design and test automation. Build cases that measure whether an agent makes a forbidden tool call or reveals a protected document. Repeat the cases, retain the execution traces, and rerun them after changing the model, permissions or workflow. A single successful demonstration does not measure how often a control fails.

AI security assurance, governance and technical leadership

Lenovo's Advisory Engineer, AI Security description combines standards and reference architectures with implementation, validation, monitoring and operational handover. Its controls cover identity, tools, data, memory, human oversight and containment.

Ripple's Senior Staff Security Engineer, AI Security advert explicitly combines securing AI systems with building AI-powered security tooling. It includes AI integration reviews, sandboxing, permission scoping, audit trails and detection of unapproved AI use.

Skills to prioritise: architecture review, technical control assurance, AI asset inventories, approval boundaries and incident response. Convert an AI-use policy into enforceable permissions, monitored integrations and a tested containment procedure. These roles also ask engineers to work across product, security and governance teams to deliver the controls.

AppSec roles that validate AI-generated findings

Caspian One's AI Application Security Engineer (Java) advert describes a different job: review AI-generated assessments, validate attack paths and false positives, inspect Java code, agree remediation with developers and track confirmed vulnerabilities to completion. Securing a model is not its stated central responsibility.

Skills to prioritise: source-code analysis, SAST triage, exploitability assessment, secure coding and remediation ownership. An experienced Java developer moving into AppSec can use existing code knowledge here, while adding security testing and vulnerability analysis. Familiarity with copilots alone does not cover the required work.

Experience requirements and realistic entry routes

These examples include substantial experience requirements. The supplied Esko description asks for at least five years in software engineering; Caspian One asks for five or more years in engineering and/or AppSec. HCLTech specifies five or more years in cybersecurity and two or more in AI security. Lenovo asks for eight or more years in related security disciplines; Ripple asks for ten or more years in security engineering, depth in at least two domains and practical AI/ML exposure. These requirements belong to the individual adverts, not every AI security vacancy.

For an existing AppSec, cloud or platform engineer, the practical move is to apply code review, identity, isolation and detection skills to AI integrations, then add agent-specific abuse tests and evaluation. For a newcomer, learn to build an authenticated API, enforce access to documents for each user, log tool actions and write tests for denied requests; then apply those controls to a small retrieval or agent application. The senior posts above still require professional experience beyond that exercise.

The seven-example source notes distinguish supplied descriptions from matching public listings, including the experience requirements. They contain paraphrased role summaries rather than a vacancy census.

Which cybersecurity jobs could shrink or disappear?

Routine Tier 1 SOC positions are a credible candidate to shrink, and a job consisting entirely of standard alert checks and escalation could disappear as a separate position in teams that automate those steps. Other exposed jobs include basic threat-intelligence compilation, scanner-report production and manual security administration. This is an assessment of task exposure, not a prediction that every employer will remove those titles.

There is already evidence of staffing pressure. The SANS/GIAC 2026 report, based on 947 respondents, records AI-related headcount reductions alongside efficiency gains. In ISC2's survey of 856 security professionals who use AI, 56% perceived reduced need for entry-level positions. That is a respondent perception, not a finding that 56% of entry-level jobs disappeared.

Tier 1 SOC analysts whose work stops at triage

Exposed work: enrich an alert with IP/domain reputation, check a standard list of events, classify an obvious false positive, write a summary and escalate. Microsoft's documented phishing-triage agent already combines message analysis, threat intelligence and hunting. Automating those steps gives an employer a route to handling a queue with fewer people, although actual staffing also depends on alert volume and required coverage.

Prioritise instead: KQL or your SIEM's query language, endpoint and identity investigation, detection tuning, and containment decisions. Be able to investigate a suspicious sign-in through privilege changes and endpoint activity, identify a missing detection, and test the revised rule. A vacancy offering investigation and detection work provides a broader development path than one confined to closing tickets.

Threat-intelligence analysts who mainly compile feeds

Exposed work: collect indicators, summarise vendor reports and produce a generic daily threat digest. ISC2 identifies reporting and basic hunting among tasks being performed or accelerated by AI tools.

Prioritise instead: assess whether a campaign's prerequisites match your organisation, map the activity to available telemetry, and turn the assessment into a tested hunt or detection. For example, check whether the affected identity platform and exposed service exist before recommending a response to a campaign report.

Pentesters who mainly run scanners and repackage results

Exposed work: run a standard scan, copy its findings and rewrite the descriptions. Scanner integration and AI-assisted finding investigation reduce the manual work in that service. This does not establish a measured decline in pentester employment.

Prioritise instead: authorisation and business-logic testing, code-assisted investigation, attack chaining and retesting fixes. Test whether a user can change another customer's order or bypass an approval workflow, rather than relying on the scanner's severity label. Those capabilities support work beyond producing a scan report.

Vulnerability coordinators limited to ticket administration

Exposed work: deduplicate findings, find owners, open tickets and send recurring reminders. These steps can be scripted through inventory, scanner and service-management APIs, with AI drafting the explanations.

Prioritise instead: affected-version checks, exposure analysis, exploitation evidence, safe mitigation and verified closure. Learn to distinguish an installed but unreachable component from an internet-exposed affected service, then check whether the deployed fix removes the actual exposure.

GRC analysts limited to copying evidence and drafting standard text

Exposed work: chase recurring screenshots, populate questionnaires and draft policy text from templates. The evidence workflow below shows how APIs and AI can reduce that administration; it does not establish that automated text proves compliance.

Prioritise instead: test whether controls operate across the stated systems and period, investigate exceptions, and assess cloud and AI access boundaries. For an access review, reconcile the privileged-account inventory to completed reviews and investigate unmatched accounts.

The career risk is narrower entry routes: employers can automate tasks that previously provided a junior analyst's first responsibilities. Ask whether a prospective SOC role includes supervised investigations, detection changes and incident work, and whether a GRC role includes control testing. Those duties provide experience beyond the tasks already targeted by automation.

Whole-field disappearance is a different claim. The US Bureau of Labor Statistics projects 21% growth in information security analyst employment from 2025 to 2035. That broader projection does not guarantee demand for a triage-only SOC position. ISC2 also found that 53% of its AI-using respondents believed AI was creating new kinds of entry-level roles. Growth and the removal of particular job designs can occur together.

Which skills and domains should you prioritise now?

ISC2's 2025 workforce study identifies cloud security, AI, security engineering, security analysis and risk assessment among hiring managers' priority technical skills. Recent UK ads also show increased shares for automation, incident response, risk management and audit-related skills. The priorities below turn those signals into specific capabilities to develop.

1. Cloud security, identity and data protection

Learn one cloud platform's human and workload identities, permission policies, temporary credentials, secure configuration and audit logs. Include the identity used by a build pipeline or AI agent, not only employee accounts. Practise finding an over-permissioned identity and restricting it without breaking the required operation.

The rationale is current hiring demand: ISC2's cloud-skills study identifies cloud security as hiring managers' leading technical priority and discusses secure design, configuration, data protection and identity/access management. The recommendation is cloud and identity competence, not memorising one provider's console.

2. AppSec and security controls inside delivery pipelines

For application security (AppSec), prioritise reading code, authentication and authorisation, API access control, threat modelling and testing a fix. Add the delivery workflow: run code, dependency and secret checks in CI; route findings to the responsible developer; rerun tests on the proposed change.

Learn how findings move between tools, including SARIF, a structured format for analysis results. The work product is a tested fix or a release control that detects a specified failure. Elucid's current advert provides a concrete example of an employer combining AppSec, CI/CD and audit responsibilities.

3. Detection, response and vulnerability management with automation

Learn to query logs, join asset and vulnerability records, enrich a finding with exposure and ownership, and automate the resulting ticket or investigation step. Python and SQL are applicable here; KQL is relevant in teams using Microsoft's hunting tools. The specific language follows the environment, but the work requires handling structured data and checking the result.

In UK core security ads, automation mentions rose from 9% in 2023 to 15% in 2025 and incident response from 10% to 18%. Sources: 2023 data, 2024 data, 2025 data.

4. Risk and control assurance with technical evidence

For GRC, prioritise risk assessment, control testing, evidence quality and exception handling. Learn how to obtain evidence from identity systems, cloud configuration, delivery pipelines and vulnerability records. For example, testing an access-review control requires the relevant identities, permissions, review period and recorded decisions; a generated policy paragraph cannot supply them.

UK risk-management mentions rose from 13% to 19% between 2023 and 2025. Auditing and ISO 27001 also increased in that series. This supports investing in technical assurance and evidence automation alongside the applicable standards.

5. AI application and agent security

If your organisation deploys AI applications, learn prompt injection, data-access boundaries, tool permissions and repeatable abuse tests. Test whether an agent can access another user's records, send data to an unapproved destination or perform a privileged action from untrusted content.

OWASP's Excessive Agency guidance gives a concrete example: an email-summarising agent also has unnecessary send permissions. The security task is to reduce the available functions and identity permissions and enforce downstream authorisation. General chatbot prompting does not cover that work.

For a technical career, combine the first two areas with automation relevant to your environment. For GRC, make technical control assurance the core capability. AI security adds a specialism where models and agents introduce an actual attack surface. These are recommendations informed by the sources, not measured job-placement probabilities.

Explore the skills and tools in job ads

Compare terms for your domain, then inspect the underlying counts and source posts. The HN tab covers ten years; the UK tab contains a separate recent job-ad series. Custom keywords, role filters and data downloads are available.

Loading the security skills explorer…

HN periods run November to October; UK observations use calendar years. Keywords may describe company context or another role in the same ad. The future-scenario control calculates a user assumption, not a hiring forecast.

What AI automation can do in everyday security work

The following are example workflows. The cited documentation establishes available capabilities; the proposed inputs, checks and outputs show how to apply them. They do not claim measured productivity gains or universal adoption. Use the organisation's approved AI environment for code, logs and customer data.

Pentesting: investigate a finding and produce reproducible evidence

Input: a scoped target, scanner finding and relevant HTTP requests and responses. Use AI to propose follow-up tests, investigate the suspected weakness and draft the report from confirmed evidence. For an access-control issue, include the two test identities and the operation each should be allowed to perform.

Check and output: reproduce the behaviour, distinguish authorised from unauthorised access and retain the actual request/response evidence. The output is a confirmed finding and retestable remediation. PortSwigger's Explore Issue with AI documents AI investigation with step logs and an HTTP logger, including sending requests to Repeater. An AI description alone does not confirm the vulnerability.

Vulnerability lifecycle management: turn findings into verified remediation

Input: scanner findings, CVE IDs, installed versions, asset exposure and service ownership. Join those records with known-exploitation evidence and the vendor's advisory. Use AI to draft a remediation order against documented rules, summarise the service impact, and propose change steps from the vendor advisory. Create the ticket through the service-management API after the asset and version checks.

Check and output: confirm that the identified version is affected, route the change to its owner, test the patch or mitigation, then rescan or verify the installed version before closing the finding. Automation can deduplicate findings and update tickets; an LLM's summary cannot establish that a patch was deployed.

CISA's KEV data records known exploited vulnerabilities. FIRST's EPSS estimates exploitation activity over the next 30 days; it is not the probability that your particular asset will be compromised. These are evidence inputs, not scores for an LLM to invent. Microsoft's Intune Vulnerability Remediation Agent documents guidance and tracking in public preview; it should not be described as proof of unattended patching or closure.

AppSec and CI/CD: produce a fix that passes security checks

Input: a code-scanning alert, affected code and relevant tests. Have the model draft a patch and a test for the original failure, then open a pull request. The pipeline reruns the scanner and the application tests; the reviewer checks that intended behaviour still works.

GitHub's agentic autofix documents investigation, CodeQL revalidation and a draft pull request. Its validation has limits, including custom queries and third-party alerts. A passed scanner check does not test every business-logic or authorisation case.

Skills to develop: scanner integration, reading security findings, writing tests for access-control failures, reviewing generated patches and maintaining merge checks. A pipeline that merely posts AI comments on every change is not the same control as a tested release gate.

Threat modelling and threat assessment: derive attack paths and control tests

Input: the current architecture, data flows, identities, sensitive assets and existing controls. Ask AI to draft abuse cases, challenge assumptions and propose tests. For a retrieval-based AI assistant, a concrete case is whether one tenant can retrieve another tenant's documents through a shared search tool.

Check and output: confirm that each proposed component and attack prerequisite exists. Test the tenant filter and downstream access check with allowed and denied requests. Record the threat, affected asset, mitigation and test result. For an advisory-based threat assessment, provide the advisory and asset/software inventory, ask which systems meet the stated attack prerequisites, then verify those matches and the relevant telemetry. Do not accept invented components, evidence or likelihood scores. OWASP SAMM's AI-assisted development guidance specifically calls for reviewing architecture context and validating generated threat and mitigation mappings.

SOC and incident response: draft a hunting query, then inspect the events

Input: an alert and a bounded question, such as which privileged-role changes followed a suspicious sign-in. Let AI draft the query and an initial timeline.

Check and output: verify tables, time ranges, joins and identity fields, run the query and inspect the returned events. Retain links to the raw records supporting the incident timeline. A query returning no rows can also indicate an incorrect field or missing data. Microsoft's advanced-hunting query assistant documents reviewing a generated query and adding it to the editor before execution. Account disabling or host isolation needs its own authorised response workflow.

Vision-language models: use screenshots in phishing triage

A vision-language model (VLM) accepts images as well as text. An example workflow combines an email or page screenshot with its actual URL, headers and security-tool results. Ask the model to identify visual impersonation indicators and relevant text, then check those observations against the message and destination.

Microsoft's Phishing Triage Agent documentation lists screenshot analysis alongside URL/file detonation, threat intelligence and hunting. That is a documented visual-analysis use case, not a claim about its underlying model architecture. A familiar logo or convincing login page does not establish sender legitimacy; a screenshot also cannot prove that a backend access control works.

GRC: collect control evidence and identify what is missing

Input: a control requirement, the systems in scope and the review period. Use scripts or APIs to collect identity settings, configuration snapshots, security-check results or remediation records. AI can draft a mapping from those records to the control and identify apparent gaps.

Check and output: confirm system coverage, timestamps, completeness and any exceptions. For an access review, a script can identify privileged accounts without a matching review record; AI can draft exception tickets from those rows with their actual record IDs. The output remains tied to a specific control and review period, with missing material still marked missing. Do not let the model turn absent evidence into a compliant status or create a citation to a record it never received.

What to prioritise for your current role

• AI security engineering: agent and tool permissions, secure retrieval, sandboxing, runtime monitoring and repeatable adversarial evaluations; use an existing software, AppSec, cloud or detection background to implement and test those controls.
• AppSec or pentesting: authentication and authorisation, API security, threat modelling and code review; then AI-assisted investigation and fixes with reproducible HTTP evidence and security tests.
• Cloud security: identity permissions, workload credentials, secure configuration and logs; then policy checks in CI/CD and automated checks for excessive permissions or exposed resources.
• Security operations or vulnerability management: log queries, asset context, exploitation evidence and remediation tracking; then AI-assisted enrichment, query generation and ticket routing with verified closure.
• GRC or risk: risk assessment, control testing and evidence collection from real systems; then automated evidence retrieval and review of the access and data boundaries of AI deployments.

Across technical roles, prioritise the programming, API and data-handling skills that connect these workflows. Learning only a dashboard's controls does not teach you how to join findings to assets, test a generated patch or identify why an automated query missed an event.

Where the work is heading

The documented tools move parts of investigation, query-writing, remediation drafting and reporting into AI-assisted workflows. Their output creates specific engineering work: checking scope, source data, permissions, exploitability and test results. The NCSC's assessment to 2027 also expects AI to accelerate parts of offensive activity and increase the attack surface.

For career development, prioritise building and operating these workflows and securing the applications that use them. The sources demonstrate capabilities and reported skill needs, not the number of jobs that AI will create or remove.

What the data shows

The HN analysis covers 120 monthly threads and 2,284 classified security hiring posts. Comparing the first and last three-year windows, AI/ML language rose from 13.7% to 34.8%, senior/staff/principal wording from 43.5% to 63.2%, teamwork/stakeholder wording from 10.4% to 20.5%, and GRC/risk from 6.0% to 10.4%. Python, AWS and penetration-testing language declined in that sample. The observations measure wording, not workplace usage or all requirements in linked descriptions.

Our priorities therefore also use independent evidence. In ISC2's 2025 workforce study, hiring managers' technical priorities included cloud security, AI, security engineering, analysis and risk assessment. In the UK series, risk management rose from 13% to 19% and incident response from 10% to 18% over 2023 to 2025. In ISC2's survey of 856 AI-using security professionals, 63% reported more time validating AI outputs.

Experience and certification requirements

In the HN sample, senior, staff or principal wording increased from 43.5% to 63.2%, while junior, graduate and internship language declined from 5.1% to 1.7%. A comment may list several roles, so these figures do not set an experience threshold for every vacancy. Elucid's 2026 advert asks for five or more years in security, AppSec or closely related software engineering, illustrating that relevant development experience can qualify for a senior security role.

Certification mentions are too sparse in the short HN comments to establish a market-wide trend. The 2020 UK report found CISSP in 37% of core security ads that requested a specific certificate, using 2016–2019 data. That conditional figure cannot be joined to the HN percentages or used to declare certifications obsolete.

The regional sources have different scopes. CyberSeek includes dedicated and adjacent US security roles; ENISA's NIS Investments research covers EU high-criticality sectors. They provide context, not a single interchangeable global skill ranking.

Conclusion

Avoid building a career around alert classification, feed summaries, scanner reports or ticket administration alone. Those are concrete automation targets. For a SOC analyst, the next priorities are investigating endpoint and identity activity, writing and tuning detections, and making evidence-based response decisions. For pentesting, move beyond scan output into authorisation, business logic and code-assisted testing.

For technical security careers, prioritise cloud identity and data protection, AppSec and secure delivery pipelines, and programming or query skills for operating security workflows. For dedicated AI security engineering roles, add agent permission controls, secure retrieval, adversarial evaluation and runtime containment. For AppSec roles using AI-generated findings, retain code analysis, exploitability assessment and remediation ownership. For GRC, prioritise risk assessment, technical control testing and automated evidence collection.

Make the learning specific: a pentest investigation that retains reproducible HTTP evidence; a vulnerability workflow that joins CVEs to exposed assets and verifies the fix; a pipeline that tests an AI-proposed patch; or a threat assessment that tests the actual access boundary.

Sources and method

We analysed security-role mentions in 120 HN hiring threads from November 2016 to October 2026 and compared them with UK reports and workforce surveys. Comments can advertise multiple roles; keyword shares do not measure all employer requirements. Chart intervals do not correct HN's selection bias. Historical ads are examples, and the learning priorities and workflows are recommendations, not placement forecasts.

Full method and limitations · HN data · UK data and sources · Reproduction scripts