GenAI security readiness lags adoption across enterprise workflows

At a glance

What this is: This is Lakera’s 2025 GenAI Security Readiness Report, which shows adoption has moved ahead of security maturity across enterprise workflows.

Why it matters: It matters because IAM, security, and AI governance teams need to close the gap between GenAI deployment and the controls that govern access, integration, and runtime behaviour.

Context

GenAI security readiness is the governance gap that appears when organisations embed generative models into workflows faster than they can define trust, access, and control boundaries. In practice, the problem is not whether teams have adopted AI, but whether they can govern what those systems can reach and influence once they are live.

Lakera’s report frames that gap from a practitioner perspective, with adoption, incidents, and capability constraints all pointing in the same direction. The article is about enterprise GenAI security maturity, not a single tool or model class, and the central question is whether current controls are keeping pace with operational use.

Key questions

Q: What breaks when GenAI is deployed without formal audit controls?

A: Without formal audit controls, organisations lose visibility into prompt behavior, context retrieval, policy enforcement, and output traceability. That means they cannot prove whether a model leaked sensitive data, ignored rules, or produced harmful content. The result is higher legal, operational, and reputational risk, especially where regulated information or customer data is involved.

Q: Why do GenAI incidents not automatically improve security readiness?

A: Incidents often raise awareness, but awareness does not create repeatable control design. Organisations may learn that GenAI is exposed to prompt injection or insecure integrations, yet still lack ownership, monitoring, and enforcement across the workflow. Maturity appears only when the response becomes operational and systematic.

Q: How should teams prioritise GenAI security investments when skills are limited?

A: Teams should prioritise governance and integration controls before expanding use cases. If skilled staff are scarce, the highest-value work is to define workflow boundaries, assign ownership, and standardise review so the organisation can secure existing deployments instead of adding more ungoverned ones.

Q: What does it mean when confidence is high but readiness is low in GenAI security?

A: It means the organisation believes its position is stronger than its operating controls actually are. In GenAI programmes, confidence can reflect policy intent or executive optimism, while readiness depends on whether teams can block unsafe prompts, contain integrations, and monitor runtime behaviour consistently.

Technical breakdown

Why GenAI readiness breaks at workflow integration

GenAI becomes harder to govern once it is embedded in customer touchpoints, developer pipelines, and internal workflows because the security boundary is no longer the model alone. The control problem shifts to the surrounding systems: prompts, connectors, data sources, and downstream actions. That creates a larger attack surface than traditional application security teams are used to handling, especially when integrations expose sensitive data or allow model outputs to influence business logic. Readiness therefore depends on more than policy statements. It depends on whether organisations can control access paths, validate outputs, and monitor behaviour in the same runtime environment where the model operates.

Practical implication: Map every GenAI workflow to the systems and data it can reach, then govern those touchpoints as part of the application control plane.

Why incidents do not automatically create maturity

The report notes that reported incidents often increase awareness without producing equivalent security improvement. That pattern is common in emerging technology adoption: organisations learn that a risk exists, but do not yet have the operational muscle to contain it. Prompt injection, insecure integrations, and manipulated outputs are not just isolated defects. They expose a lack of structured defensive design around how the model receives instructions, consumes data, and returns output. In other words, the organisation reacts to the event, but the governance model remains unchanged. Maturity appears when the response becomes systematic rather than anecdotal.

Practical implication: Treat each incident as evidence of a repeatable control gap and convert it into a standing governance requirement, not a one-off fix.

The compound readiness gap is a workforce and architecture problem

Lakera’s analysis points to two constraints that reinforce each other: shortage of skilled AI-security talent and the complexity of integrating GenAI into existing infrastructure. That combination creates a compound readiness gap, where the organisation may want to secure AI but cannot yet operationalise the controls. This is not simply a staffing issue. It is a programme design issue, because AI governance, IAM, and security engineering are being forced to converge faster than many teams can absorb. The result is delayed control implementation, inconsistent review, and unclear ownership across teams.

Practical implication: Assign explicit ownership for GenAI security governance across security, platform, and application teams before scaling production use.

Threat narrative

Attacker objective: The objective is to influence or exploit GenAI workflows so that sensitive data, system behaviour, or decision outputs can be manipulated at operational scale.

1. Entry occurs through prompt injection, insecure integrations, or manipulated inputs that enter the GenAI workflow through legitimate application paths.
2. Credential or data access is abused when the model or connected system can reach information sources or actions beyond the intended trust boundary.
3. Escalation follows when the output or toolchain is used to override system instructions, leak data, or produce harmful actions at scale.
4. Impact is realised through exposed data, compromised decision flows, or security controls that fail to keep pace with runtime misuse.

Breaches seen in the wild

• iOS apps leaking hard-coded secrets: Cybernews found 71% of 156,080 iOS apps leak hard-coded secrets, with open cloud storage and Firebase databases exposing user data.

Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

Readiness is now a runtime governance problem, not a model-adoption problem. Once GenAI is embedded in live workflows, the meaningful question shifts from whether the organisation is using AI to whether it can govern what the system can access and do in real time. That distinction matters because workflow integration expands the control surface beyond the model itself. Practitioners should treat GenAI security as part of operational identity and access governance, not a separate innovation track.

Capability gaps matter more than confidence scores. The report’s confidence-versus-readiness split shows that organisations can believe they are prepared while still lacking the controls to manage prompt injection, insecure integrations, and output manipulation. Confidence is a perception metric, but governance requires repeatable control enforcement. The practitioner lesson is that readiness must be tested in production conditions, not inferred from policy existence.

Compound readiness gap: the shortage of AI-security talent and the complexity of integrating GenAI into legacy environments create a governance bottleneck that no single team can solve alone. Security programmes that keep AI, IAM, and application ownership separate will remain slow to act and inconsistent in execution. The field is moving toward cross-functional identity and security governance for AI, and that shift is already underway.

Incident learning is not enough when the attack surface keeps changing. The article shows that awareness rises after incidents, but structured defense does not automatically follow. That pattern echoes earlier cloud adoption, where visibility improved before governance did. The implication is that GenAI programmes need controls that operate continuously, because runtime attacks evolve faster than review cycles.

Privacy is no longer the only headline risk. The report shows concern shifting toward adversarial misuse, agent risks, and offensive AI, which means GenAI security is broadening from data protection into behaviour and action control. For identity practitioners, that is a clear signal that access governance, tool permissions, and runtime monitoring are becoming central to AI security programmes.

From our research library:

• Generative AI use specifically increased from 33% in 2023 to 79% in 2025, according to McKinsey’s Global Surveys on the State of AI.

What this signals

GenAI readiness now depends on governance at the workflow edge. The decisive control question is no longer whether the model is approved, but whether the organisation can constrain prompts, connectors, and actions once the system is live. Teams that keep treating GenAI as a point solution will miss the fact that the risk sits in the surrounding workflow.

Adoption without operating discipline creates security debt. When GenAI is rolled into production faster than ownership, review, and monitoring can scale, the result is a growing backlog of exceptions and unmanaged behaviour. Security programmes should expect that backlog to surface first in integration-heavy use cases, where model output can directly affect business processes.

For practitioners

• Define GenAI workflow trust boundaries Inventory every model, connector, and downstream action path so teams know which data sources and tools each GenAI workflow can reach.
• Separate prompt, data, and action controls Use distinct controls for what the model can read, what it can reference, and what it can trigger so a single failure does not cascade across the workflow.
• Track incident classes as governance signals Classify prompt injection, insecure integrations, and manipulated outputs as repeatable governance failures rather than isolated incidents.
• Assign cross-functional ownership Make security, platform, and application teams jointly accountable for GenAI control design, review, and exception handling.

Key takeaways

• GenAI adoption is now widespread enough that governance gaps are no longer hypothetical, and the report shows readiness has not kept pace with deployment.
• The strongest evidence of strain is the split between confidence and concern, which signals that many organisations know the risk exists but lack consistent control coverage.
• The practical response is to govern GenAI as a workflow problem, with explicit ownership, boundary control, and runtime monitoring across every connected system.

Key terms

• GenAI security readiness: The extent to which an organisation can safely deploy generative AI in production without losing control over data, access, or downstream actions. It combines policy, architecture, skills, and operational testing, because a secure model is not enough if the surrounding workflow remains open to abuse.
• Prompt Injection (Agentic): An attack where malicious instructions are embedded in content that an AI agent reads, causing the agent to execute unintended actions using its own legitimate credentials. A primary vector for agent goal hijacking and identity abuse.
• Workflow trust boundary: A workflow trust boundary is the point where one step in a business or technical process stops being fully trusted and must be checked before the next step continues. It defines where identity, data, or action permissions change, and where controls such as approval, validation, logging, or isolation should be applied.
• Runtime Confrontation: Runtime confrontation describes security conditions where attacks and misuse happen while the AI system is active, not just during build or review stages. It requires controls that observe and constrain live behaviour, especially when outputs can trigger real-world actions.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity lifecycle are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are responsible for identity security strategy or NHI governance in your organisation, it is worth exploring.