Secure and Responsible AI
Employees use AI chats, teams build their own assistants, and AI agents gain access to data and systems. We help you do it securely: policies and risk assessment from SafeWave, data and human protection in AI use with Proofpoint, application and AI agent protection, and employee AI use control with Check Point.
13%
of organizations reported a breach of AI models or applications – and 97% of them lacked AI access control. This is according to the IBM Cost of a Data Breach 2025.1
63%
of organizations that suffered a breach either have no AI governance policy or are just developing one.1
$670K
– this was the average higher cost of a breach in companies with a high scale of unauthorized AI use (“shadow AI”).1
16%
of breaches involved attackers using AI – most often for phishing and impersonation (deepfake).1
2027
AI Act requirements for high-risk systems from Annex III become effective from December 2 – after postponement by the Digital Omnibus.2
Conclusion: the problem is not AI itself, but the lack of rules and control around it. In our opinion, prohibition doesn’t work – employees will use the tools anyway. Visibility, policy, and the technology that enforces it do work.
Scenarios from the IBM report, OWASP Top 10 for LLM Applications list, and vendor materials. Risk applies to both how people use AI and what the company builds with AI itself. For each: what technology or our team does in this area.
A contract, code, or customer list pasted into a free AI tool. In an IBM study, one in five organizations experienced a breach related to unauthorized AI use.1
Proofpoint AI Data Governance: shows approved and unapproved AI tools in the company and applies ready-made policies against data leaks and privacy breaches.3
An attacker hides an instruction in the message content that an AI assistant reading the email will execute. Prompt injection tops the OWASP Top 10 for LLM Applications list.4
Proofpoint Prime Threat Protection: according to the vendor, it detects and blocks malicious commands hidden in emails, targeting AI assistants.3
An assistant for customers or employees, built on a language model, reveals data or breaks rules after a cleverly phrased question.
Check Point (Lakera): runtime protection for AI models, applications, and agents – against prompt injection, data leakage, and model manipulation.5
The agent has access to email, files, and systems – and will do more than it should. “Excessive agency” is one of the OWASP list items.4
Proofpoint Secure Agent Gateway: controls how AI agents access data and blocks or masks sensitive data. The vendor is rolling it out in phases starting in 2026.3
A new AI feature reaches customers without checking how it reacts to abuse attempts – because application security tests do not cover models.
Check Point (Lakera) AI Red Teaming: pre-deployment assessment and continuous testing, powered by a database of over 80 million attack patterns.5
63% of organizations after a breach had no AI governance policy.1 Meanwhile, transparency obligations from the AI Act, among others, become effective from August 2, 2026.2
SafeWave: AI use case inventory, usage policy, risk assessment, and classification according to the AI Act – with reference to ISO/IEC 42001.6
Many companies already have a license for an AI assistant and a ban on pasting data in their terms of service. The problem is that terms of service don’t see what’s happening.
Blocking one chat site shifts the problem to the next. Control must apply to data, not addresses.
SafeWave
Secure AI starts with decisions, not tools: what is allowed, by whom, with what data, and who is responsible for the risk. We help make and document these decisions, then enforce them with technology.
What AI tools and models are in use – approved and unapproved – and what data they access.
Clear rules for employees and teams: what is allowed, with what data, and in which tools.
Risk of each use case: data, decisions, people, model provider.
Classification of use cases and obligations – transparency from August 2026, high-risk systems from December 2027.2
Roles, processes, and reviews in the spirit of ISO/IEC 42001 for AI management systems.6
Check Point for applications, agents, and employees using AI, Proofpoint for data – implemented and fine-tuned by us.
We do not sell a single AI model or platform. As a vendor-agnostic integrator, we advise on what makes sense for your situation.
Check Point (Lakera)
98%+ detection effectiveness with latency below 50 ms – according to the vendor.5
Proofpoint
10,000+ large companies use Proofpoint – according to the vendor.3
Check Point Workforce AI Security
100+ languages monitored in real-time – according to the vendor.8
The AI Act did not become effective on a single day. Obligations are introduced in stages, and in July 2026, some deadlines were shifted by the Digital Omnibus regulation.2
Transparency, among other things, is already in force: AI systems that interact with humans must generally inform that it is AI. The scope of obligations depends on the company’s role – provider or deployer – and the system’s risk level.9
Regulation (EU) 2024/1689 establishes obligations based on the role and risk level of the AI system.9 Deadlines for high-risk systems were changed by Regulation (EU) 2026/1744.2
Processing personal data by AI requires a legal basis, and for high risk – a data protection impact assessment (Article 35). Article 22 restricts decisions based solely on automated processing.10
Regulations do not specify products. We help combine them into one coherent AI governance framework – instead of three separate compliance projects.
SafeWave has been operating since 1998 as a vendor-agnostic integrator – also in AI. We advise on what makes sense for your situation, not just one product.
We determine which AI tools the company uses, by whom, and with what data – even beyond official licenses.
AI usage policy, risk assessment of use cases, and classification according to the AI Act.
Data control in AI use with Proofpoint and protection of proprietary AI applications and agents with Check Point.
Regular review of new tools and use cases, training for employees and management. Support in English and Polish.
Let’s start with an inventory of AI tools and a brief risk assessment. You will receive a list of use cases, risks, and initial policies to implement.