# Bezpieczeństwo i governance

EN: **AI safety, security and governance** · id `safety`

Path: [AI Atlas](https://aiatlasgraph.com/llms.txt) › Bezpieczeństwo i governance

## Description

- PL: Ryzyko, zabezpieczenia i odpowiedzialność za system AI.
- EN: Risk, safeguards and responsibility for an AI system.

## In one sentence

- PL: To ograniczanie szkód powodowanych przez modele oraz ustalanie, kto odpowiada za ich działanie.
- EN: This limits harm from models and establishes who is responsible for their operation.

## Problem it solves

- PL: Model może ujawnić dane, potraktować ludzi nierówno lub wykonać szkodliwe działanie, mimo że zwykle odpowiada poprawnie.
- EN: A model can reveal data, treat people unequally or perform harmful actions even when its responses are usually correct.

## What it does

- PL: Łączy ocenę ryzyka, trening, ochronę danych, ograniczenia działania i nadzór; skuteczność sprawdza się również po udostępnieniu systemu.
- EN: It combines risk assessment, training, data protection, action limits and oversight; effectiveness is also checked after a system is released.

## Everyday example

- PL: Przed uruchomieniem asystenta sprawdzasz ujawnianie danych, dostęp do narzędzi i sposób zgłaszania błędów przez użytkowników.
- EN: Before launching an assistant, you check data exposure, tool access and how users can report mistakes.

## This concerns you if you…

- PL: udostępniasz model innym osobom
- PL: jego odpowiedzi lub działania mogą komuś zaszkodzić
- EN: make a model available to other people
- EN: use outputs or actions that could harm someone

## Keywords

NIST AI RMF · OWASP LLM Top 10 · EU AI Act · RLHF · Constitutional AI · AI red teaming · differential privacy · Fairlearn · C2PA · Llama Guard · model cards · responsible scaling policies · prompt injection

## How it works

- PL: Model zagrożeń łączy możliwości systemu, dostępne dane i uprawnienia z możliwymi szkodami. Ewaluacje, ograniczenia wykonania oraz proces odpowiedzialności tworzą warstwy kontroli, aktualizowane na podstawie incydentów i zmian systemu.
- EN: A threat model connects system capabilities, available data and permissions to possible harms. Evaluations, execution constraints and accountability processes form control layers updated from incidents and system changes.

## Limits

- PL: Żaden pojedynczy test lub filtr nie obejmuje wszystkich szkód. Zabezpieczenia zależą od środowiska i zakresu uprawnień; zmiany modelu, narzędzi lub użytkowników wymagają ponownej oceny.
- EN: No single test or filter covers every harm. Controls depend on environments and permission scope; model, tool or user changes require reassessment.

## Key terms

- PL: **model zagrożeń**: Opis chronionych zasobów, przeciwników i możliwych dróg szkody.
- PL: **kontrola warstwowa**: Łączenie zabezpieczeń działających na różnych etapach i granicach systemu.
- EN: **threat model**: A description of protected assets, adversaries and potential harm pathways.
- EN: **layered control**: Combining safeguards across different system stages and boundaries.

## Know first

- [Zbiory testowe i regresje (Evaluation datasets and regression tests)](https://aiatlasgraph.com/t/evalsets.md)
- [Workflow i udział człowieka (Workflows and human-in-the-loop)](https://aiatlasgraph.com/t/workflows.md)

## Examples

- PL: Alignment · red teaming · prompt injection · prywatność · AI Act
- EN: Alignment · red teaming · prompt injection · privacy · AI Act

## Subtopics (8 direct, 40 in total)

- [Alignment i safety (Alignment and safety)](https://aiatlasgraph.com/t/alignment.md): Aligning model behaviour with intended goals and limits: how to train it, how to oversee models stronger than humans, and how to detect a model gaming its objective or hiding its intent.
  - [Obchodzenie celu: reward hacking i służalczość (Reward hacking, specification gaming and sycophancy)](https://aiatlasgraph.com/t/rewardhacking.md): The model satisfies the literal training objective but not its intent: it exploits loopholes in the reward (e.g. gaming tests) or tells users what they want to hear instead of the truth. Detecting and preventing this is a practical problem in RL training today.
    - [Lokalizacja niepożądanych zachowań przez routowanie gradientów (Gradient-routed localization of unintended behavior)](https://aiatlasgraph.com/t/graft.md): Localizes reward hacking in a removable adapter by routing gradients.
  - [Postępujące rozminięcie z celami (Progressive misalignment)](https://aiatlasgraph.com/t/progressivemisalignment.md): An observed phenomenon in which a model increasingly answers harmful queries across rounds of recursive self-distillation despite benign extra information for the teacher.
  - [Skalowalny nadzór (debata, weak-to-strong) (Scalable oversight)](https://aiatlasgraph.com/t/scalableoversight.md): Methods for supervising models that may outperform humans on a task: humans assisted by an AI, a debate between two models judged by a human, a weaker model supervising a stronger one.
  - [Trening według zasad (Constitutional AI, RLAIF) (Constitutional AI and RL from AI feedback)](https://aiatlasgraph.com/t/constitutionalai.md): Training a model to be harmless from a written list of principles (a “constitution”): the model critiques and revises its own answers, and preference labels come from an AI instead of humans (RLAIF).
  - [Udawane dopasowanie i ukryte cele (alignment faking, scheming) (Alignment faking and scheming)](https://aiatlasgraph.com/t/scheming.md): Studying whether a model can strategically hide its goals: behaving well only when it believes it is being trained or tested, deliberately underperforming (sandbagging) or evading oversight. An evaluation research area since 2024.
- [Bezpieczeństwo aplikacji AI (AI application security)](https://aiatlasgraph.com/t/security.md): Protecting data, tools and execution.
  - [Ataki przez prompt (Prompt attacks)](https://aiatlasgraph.com/t/promptattacks.md): Hijacking model behaviour through input content: injected instructions, leaking the prompt or data, bypassing safeguards.
    - [Jailbreaki — obchodzenie zabezpieczeń modelu (Jailbreaking)](https://aiatlasgraph.com/t/jailbreaks.md): Prompts that get a model to break its own safety rules: role-play, obfuscated requests, low-resource languages, automatically searched adversarial suffixes. They work because the goals “be helpful” and “be safe” compete.
    - [Pośrednie wstrzykiwanie promptu (przez dokumenty i narzędzia) (Indirect prompt injection)](https://aiatlasgraph.com/t/indirectinjection.md): The attacker never talks to the model — they hide instructions in a web page, email, file or tool result that the LLM application will fetch itself. Especially dangerous for RAG and agents, as it can lead to data theft or unwanted actions.
  - [Bezpieczeństwo agentów: uprawnienia i izolacja (AI agent security)](https://aiatlasgraph.com/t/agentsecurity.md): Designing agents so that a compromised model cannot cause harm: least privilege, sandboxed code execution, separating trusted instructions from untrusted data, confirmations before risky actions.
    - [Mechanistyczne sterowanie bezpieczeństwem agentów CLI (Mechanistic subspace safety steering for multi-turn agents)](https://aiatlasgraph.com/t/mechsuits.md): An experimental white-box method detects harmful execution states of multi-turn CLI agents from hidden representations and intervenes in ten selected dimensions of a single layer.
    - [Monitor interwencji krokowej dla agentów (Step-level timely intervention monitors for agents)](https://aiatlasgraph.com/t/stepsafe.md): A step-level reasoning guardrail that chooses CONTINUE, ASK or STOP before each agent action and is trained with Adaptive Stage-Aligned Policy Optimization (ASPO).
  - [Bezpieczeństwo łańcucha dostaw modeli (AI model supply chain security)](https://aiatlasgraph.com/t/supplychain.md): Risks of pulling in third-party weights, adapters, datasets and libraries: malicious code in model files, hidden backdoors, unknown provenance. Practice: safe weight formats, source verification, scanning.
  - [Zatruwanie danych i backdoory (Data poisoning and backdoors)](https://aiatlasgraph.com/t/poisoning.md): Attacking the training data: false information or a hidden trigger that changes the model's behaviour.
    - [Podgląd aktualizacji dla bezpiecznego dostrajania (Update-preview defense for safe fine-tuning)](https://aiatlasgraph.com/t/lookaheaddefense.md): Previews and penalizes fine-tuning updates that weaken safe behaviour from a buffer.
- [Ewaluacje bezpieczeństwa modeli (AI safety evaluations)](https://aiatlasgraph.com/t/safetyevals.md): Deliberately probing a model for harmful behaviour and dangerous capabilities before and after deployment — by people, other models and standardised tests.
  - [Ewaluacje niebezpiecznych zdolności (cyber, bio, autonomia) (Dangerous capability evaluations)](https://aiatlasgraph.com/t/dangerouscapevals.md): Tests of whether a model can meaningfully help with dangerous activities — cyberattacks, biological or chemical weapons, persuasion, self-proliferation or automating AI research. Results trigger thresholds in companies' safety policies.
  - [Red teaming modeli (ręczny i automatyczny) (Red teaming language models)](https://aiatlasgraph.com/t/redteaming.md): Attacking your own model to find harmful outputs before users do: manually by tester teams, or automatically with one model generating attacks against another.
  - [Samocena rozmowy pod kątem nadużyć (Conversation self-assessment for misuse detection)](https://aiatlasgraph.com/t/selfassessment.md): After responding, the model generates a short safety assessment of its own conversation, reusing the already computed context.
  - [Sondy aktywacji do wykrywania niepokojących wzorców (Activation probes for detecting concerning model behavior)](https://aiatlasgraph.com/t/activationprobesafety.md): Lightweight classifiers analysing the model's internal activations for patterns associated with undesirable behaviour.
  - [Wielowarstwowe monitorowanie agentów (Multi-layer monitoring for agent misalignment)](https://aiatlasgraph.com/t/multilayeragentmonitor.md): A cascade of monitors from cheap activation probes, through model reviewers, to investigative agents analysing full transcripts and logs.
- [Governance i zarządzanie ryzykiem (Governance and risk management)](https://aiatlasgraph.com/t/governance.md): Accountability, policies, documentation and incidents.
  - [Dokumentacja modeli i danych (karty modeli) (Model and dataset documentation)](https://aiatlasgraph.com/t/modeldocs.md): Short, standardised documents shipped with a model or dataset: intended use, limitations, evaluation procedure, results across groups, data origin and composition.
  - [Polityki bezpieczeństwa modeli granicznych (Frontier AI safety policies)](https://aiatlasgraph.com/t/frontierframeworks.md): Public commitments by companies building the most capable models: which dangerous-capability thresholds they test for, which safeguards then apply, and when they would pause deployment or development.
  - [Ramy zarządzania ryzykiem AI (NIST AI RMF, ISO/IEC 42001) (AI risk management frameworks)](https://aiatlasgraph.com/t/riskframeworks.md): Voluntary frameworks and standards for organising AI risk management: who is accountable, how to map and measure risks, how to respond. ISO/IEC 42001 specifies an organisational AI management system.
  - [Regulacje AI i klasy ryzyka (AI Act) (AI regulation and risk tiers (EU AI Act))](https://aiatlasgraph.com/t/airegulation.md): Law that sorts AI systems by risk level — prohibited practices, high risk, transparency obligations, minimal risk — with separate duties for general-purpose model providers. In the EU the rules phase in over 2025–2028.
- [Guardrails (Guardrails and output control)](https://aiatlasgraph.com/t/guardrails.md): Blocks and filters on the model's input and output.
  - [Klasyfikatory bezpieczeństwa wejścia i wyjścia (Input and output safety classifiers)](https://aiatlasgraph.com/t/safetyclassifiers.md): Separate, usually smaller models that score the prompt and the response against a harm taxonomy and block what is not allowed. They can be adapted to a custom category list.
  - [Programowalne reguły przepływu rozmowy (rails) (Programmable dialogue rails)](https://aiatlasgraph.com/t/programmablerails.md): Rules plugged into different stages of an LLM application — input, dialogue, retrieval, tool execution and output — that keep the conversation on topic and on script.
- [Prywatność i prawa do danych (Privacy and data rights)](https://aiatlasgraph.com/t/privacy.md): Rules for using data and models.
  - [Oduczanie modeli (machine unlearning) (Machine unlearning)](https://aiatlasgraph.com/t/unlearning.md): Removing the influence of selected data or capabilities (e.g. personal data, copyrighted content) from an already-trained model without retraining from scratch. Still an open problem: it is hard to verify the model truly “forgot”.
  - [Prywatność różnicowa (DP-SGD) (Differential privacy in machine learning)](https://aiatlasgraph.com/t/diffprivacy.md): A mathematical guarantee that the training result barely depends on whether any single person is in the data. In practice: gradient clipping and added noise (DP-SGD) under a controlled “privacy budget” — at a cost to model quality.
  - [Uczenie federacyjne (Federated learning)](https://aiatlasgraph.com/t/federated.md): Collaborative model training without centralising raw data: data stays on participants’ devices or servers, which send updates for aggregation. Federated learning alone does not guarantee privacy — updates can leak information, so secure aggregation and differential privacy are among the additional protections used.
  - [Wykrywanie i anonimizacja danych osobowych (PII) (PII detection and anonymisation)](https://aiatlasgraph.com/t/piiredaction.md): Automatically finding personal data (names, numbers, addresses) in text, images and tables and masking or replacing it — before training, in logs and in prompts sent to models. No tool catches everything.
- [Stronniczość i sprawiedliwość (Bias and fairness)](https://aiatlasgraph.com/t/fairness.md): Effects of a system behaving differently toward different groups.
  - [Metody ograniczania stronniczości (Bias mitigation techniques)](https://aiatlasgraph.com/t/biasmitigation.md): Reducing disparities at three stages: before training (transforming data), during training (fairness-constrained optimisation) and after training (e.g. group-specific decision thresholds).
  - [Metryki sprawiedliwości grupowej (Group fairness metrics)](https://aiatlasgraph.com/t/fairnessmetrics.md): Comparing model behaviour across groups defined by sensitive features: are positive decisions and errors distributed similarly? Different metrics conflict with each other, and a near-zero gap does not by itself mean the result is fair.
- [Znaki wodne i pochodzenie treści AI (AI content watermarking and provenance)](https://aiatlasgraph.com/t/contentprovenance.md): Techniques that signal content origin: invisible watermarks embedded during generation and cryptographically signed metadata manifests. Their limitations differ: editing and paraphrasing can weaken watermarks, while C2PA metadata can be detached from the file.

## Related topics

Untyped, symmetric "see also" links.

- [AI w cyberbezpieczeństwie (AI for cybersecurity)](https://aiatlasgraph.com/t/cyberai.md)

## Sources

- [NIST · AI Risk Management Framework](<https://www.nist.gov/itl/ai-risk-management-framework>) `nist` · reference: Uzupełnienie kursów o ryzyko i zarządzanie AI; nie porada prawna.
- [Full Stack Deep Learning (2022, archiwum)](<https://fullstackdeeplearning.com/course/2022/>) `fsdl` · reference: Pełny cykl systemu ML: dane, testy, wdrażanie, monitoring, zespoły i etyka.
- [International AI Safety Report 2026](<https://internationalaisafetyreport.org/publication/international-ai-safety-report-2026>) `international-ai-safety-report-2026` · by Panel ponad 100 ekspertów pod przewodnictwem Yoshuy Bengio; Expert Advisory Panel z przedstawicielami 30+ państw i organizacji

---

AI Atlas revision 122 (2026-10-30) · HTML: https://aiatlasgraph.com/t/safety · index: https://aiatlasgraph.com/llms.txt · everything: https://aiatlasgraph.com/llms-full.txt · updated 2026-10-30