The New Baseline for AI Safety
AI is now not an experimental functionality or a back-office automation device: it’s changing into a core operational layer inside trendy enterprises. The tempo of adoption is breathtaking. But, in response to Cisco’s 2025 AI Readiness Indexsolely 29 p.c of firms imagine they’re adequately geared up to defend towards AI threats and solely 33 p.c have a proper change-management plan for guiding accountable adoption.
Executives and leaders more and more discover themselves in a troubling place: they perceive cybersecurity, however AI safety feels international. People, organizations, and governments can’t adequately comprehend or reply to the implications of such quickly evolving know-how and the threats that ensue: organizations are deploying techniques whose conduct evolves, whose modes of failure aren’t absolutely understood, and whose interactions with their setting are dynamic and typically unpredictable.
Cisco’s Built-in AI Safety and Security Framework (additionally referred to on this weblog as “AI Safety Framework”) provides a basically completely different method. It represents one of many first holistic makes an attempt to categorise, combine, and operationalize the complete vary of AI dangers, from adversarial threats, content material security failures, mannequin and provide chain compromise, agentic behaviors and ecosystem dangers (e.g., orchestration abuse, multi-agent collusion), and organizational governance. This vendor-agnostic framework gives a construction for understanding how trendy AI techniques fail, how adversaries exploit them, and the way organizations can construct defenses that evolve alongside functionality developments.
A Fragmented Panorama—and the Want for Integration
For years, organizations that tried to safe AI pieced collectively steerage from disparate sources. MITRE ATLAS helped outline adversarial techniques in machine studying techniques. NIST’s Adversarial Machine Studying taxonomy described assault primitives. OWASP printed High 10 lists for LLM and agentic dangers. Frontier AI labs like Google, OpenAI, and Anthropic shared inside security practices and rules. But every of those efforts centered on a specific slice of the danger panorama, providing items of the puzzle however cease in need of offering a unified, end-to-end understanding of AI threat.
What has been lacking is a cohesive mannequin—one which seamlessly spans security and safety, runtime and provide chain, mannequin conduct and system conduct, enter manipulation and dangerous outputs. Cisco’s evaluation makes the hole clear: no present framework covers content material harms, agentic dangers, provide chain threats, multimodal vulnerabilities, and lifecycle-level publicity with the completeness wanted for enterprise-grade deployment. The actual world doesn’t phase these domains, and adversaries definitely don’t both.
Evaluation of protection throughout AI safety taxonomies and frameworks
A New Paradigm for Understanding AI Threat
AI safety and security dangers current very actual issues for organizations. Taken collectively, AI safety and AI security type complementary dimensions of a unified threat framework: one involved with defending AI techniques from threats, and the opposite with making certain that their conduct stays aligned with human values and ethics. Treating these domains in tandem can allow organizations to construct AI techniques that aren’t solely strong and dependable, but additionally accountable and worthy of belief.
We outline them as:
- AI safety: the self-discipline of making certain AI accountability and defending AI techniques from unauthorized use, availability assaults, and integrity compromise throughout the AI lifecycle.
- AI security: serving to guarantee AI techniques behave ethically, reliably, pretty, transparently, and in alignment with human values.
Cisco’s Built-in AI Safety and Security Framework is constructed upon 5 design parts that distinguish it from prior taxonomic efforts and embody an evolving AI risk panorama: the combination of AI threats and content material harms, AI improvement lifecycle consciousness, multi-agent coordination, multimodality, and audience-aware utility.
(1) Integration of threats and harms: One core innovation of Cisco’s framework is its recognition that AI safety and AI security are inseparable. Adversaries exploit vulnerabilities throughout each domains, and oftentimes, hyperlink content material manipulation with technical exploits to attain their targets. A safety assault, equivalent to injecting malicious directions or corrupting coaching knowledge, typically culminates in a security failure, equivalent to producing dangerous content material, leaking confidential info, or producing undesirable or dangerous outputs.
Conventional approaches have handled security and safety as parallel tracks. Our AI Safety Framework makes an attempt to replicate the fact of contemporary AI techniques: the place adversarial conduct, supposed and unintended system conduct, and person hurt are interconnected. The AI Safety Framework’s taxonomy brings these parts right into a single construction that organizations can use to grasp threat holistically and construct defenses that deal with each the mechanism of assault and the ensuing affect.
(2) AI lifecycle consciousness: One other defining characteristic of the AI Safety Framework is its anchor within the full AI lifecycle. Safety issues throughout knowledge assortment and preprocessing differ from these throughout mannequin coaching, deployment and integration, device use, or runtime operation. Vulnerabilities which can be irrelevant throughout mannequin improvement might turn into crucial as soon as the mannequin features entry to tooling or interacts with different brokers. Our AI Safety Framework follows the mannequin throughout this complete journey, making it clear the place completely different classes of threat emerge and the way they might evolve, and permitting organizations to implement defense-in-depth methods that account for the way dangers evolve as AI techniques progress from improvement to manufacturing.
(3) Multi-agent orchestration: The AI Safety Framework may also account for the dangers that emerge when AI techniques work collectively, encompassing orchestration patterns, inter-agent communication protocols, shared reminiscence architectures, and collaborative decision-making processes. Our taxonomy accounts for related dangers that emerge in techniques with autonomous planning capabilities (brokers), exterior device entry (MCP1), persistent reminiscence, and multi-agent collaboration—threats that might be invisible to frameworks designed for earlier generations of AI know-how.
(4) Multimodality issues: The AI Safety Framework additionally displays the fact that AI is more and more multimodal. Threats can emerge from textual content prompts, audio instructions, maliciously constructed photographs, manipulated video, corrupted code snippets, and even embedded alerts in sensor knowledge. As we proceed to analysis how multimodal threats can manifest, treating these pathways persistently is crucial, particularly as organizations undertake multimodal techniques in robotics and autonomous automobile deployments, buyer expertise platforms, and real-time monitoring environments.
(5) An audience-aware safety compass: Lastly, the framework is deliberately designed for a number of audiences. Executives can function on the degree of attacker targets: broad classes of threat that map on to enterprise publicity, regulatory issues, and reputational affect. Safety leaders can give attention to strategies, whereas engineers and researchers can dive deeper into subtechniques. Drilling down even additional, AI purple groups and risk intelligence groups can construct, check, and consider procedures. All of those teams can share a single conceptual mannequin, creating alignment that has been lacking from the trade.
The AI Safety Framework gives groups with a shared language and psychological mannequin for understanding the risk panorama past particular person mannequin architectures. The framework consists of the supporting infrastructure, complicated provide chains, organizational insurance policies, and human-in-the-loop interactions that collectively decide safety outcomes. This allows clearer communication between AI builders, AI end-users, enterprise features, safety practitioners, and governance and compliance entities.
Contained in the AI Safety Framework: A Unified Taxonomy of AI Threats
A vital element of the AI Safety Framework is the underlying taxonomy of AI threats that’s structured into 4 layers: targets (the “why” behind assaults), strategies (the “how”), subtechniques (particular variants of “how”), and procedures (real-world implementations). This hierarchy creates a logical, traceable pathway from high-level motivations to detailed implementation.
The framework identifies nineteen attacker targets, starting from purpose hijacking and jailbreaks to communication compromise, knowledge privateness violations, privilege escalation, dangerous content material era, and cyber-physical manipulation. These targets map on to noticed patterns and threats, to vulnerabilities organizations are encountering as they scale AI adoption, and at last lengthen to areas which can be technically possible, although not but noticed outdoors of a analysis setting. Every goal turns into a lens by way of which executives and leaders can perceive their publicity: which enterprise features may very well be impacted, which regulatory obligations could be triggered, and which techniques require heightened monitoring.
Strategies and subtechniques present the specificity mandatory for operational groups. These embrace over 150 strategies and subtechniques equivalent to immediate injections (each direct and oblique), jailbreaks, multi-agent manipulation, reminiscence corruption, provide chain tampering, environment-aware evasion, device exploitation, and dozens extra. The richness of this layer displays the complexity of contemporary AI ecosystems. A single malicious immediate might propagate throughout brokers, instruments, reminiscence shops, and APIs; a single compromised dependency might introduce unobserved backdoors into mannequin weights; or a single cascaded failure might trigger a whole multi-agent workflow to diverge from its supposed purpose.
Screenshot of the AI Safety Framework’s Taxonomy Navigator
The protection taxonomy embedded throughout the framework is equally strong. It consists of twenty-five classes of dangerous content material, starting from cybersecurity misuse to security and content material harms to mental property compromise and privateness assaults. This breadth acknowledges that many AI failures are emergent behaviors that may nonetheless trigger real-world hurt. A unified taxonomy ensures that organizations can consider each malicious inputs and dangerous outputs by way of a coherent lens.
Alongside that vein, there are extra mannequin context protocol (MCP), agentic, and provide chain risk taxonomies embedded throughout the AI Safety Framework. Protocols like MCP and A2A govern how LLMs interpret instruments, prompts, metadata, and execution environments, and when these elements are tampered with, impersonated, or misused, benign agent operations could be redirected towards malicious targets. The MCP taxonomy (which presently covers 14 risk varieties) and our A2A taxonomy (which presently covers 17 risk varieties) are each standalone assets which can be additionally built-in into AI Protection and in our open supply instruments: MCP Scanner and A2A Scanner. Lastly, provide chain threat can be a core dimension of lifecycle-aware AI safety. We’ve developed a taxonomy that covers 22 distinct threats and is equally built-in into AI Protection, our companions in mannequin safety, and different instruments we’re creating for the open supply group.
Cisco’s Built-in AI Safety and Security Framework provides one of the full, forward-looking approaches out there in the present day. At a time when AI is redefining industries, that readability will not be merely beneficial—it’s important. This framework can be built-in into Cisco AI Protection, the place threats are recognized with related indicators and mitigation methods. Navigate our Built-in AI Safety and Security Framework in the present day. We look ahead to working with the group to deepen the attention and strengthen defenses towards this novel ecosystem of AI threats.
