The stakes of one thing going improper with AI are extremely excessive. Solely 29% of organizations really feel totally geared up to detect and stop unauthorized tampering with AI[1]. With AI, rising dangers goal totally different phases of the AI lifecycle, whereas accountability lies with totally different homeowners together with builders, finish customers and distributors.
As AI turns into ubiquitous, enterprises will use and develop a whole lot if not 1000’s of AI purposes. Builders want AI safety and security guardrails that work for each utility. In parallel, deployers and finish customers are dashing to undertake AI to enhance productiveness, probably exposing their group to information leakage or the poisoning of proprietary information. This provides to the rising dangers associated to organizations shifting past public information to coach fashions on their proprietary information.
So, how can we make sure the safety of AI techniques? Methods to shield AI from unauthorized entry and misuse? Or forestall information from leaking? Guaranteeing the safety and moral use of AI techniques has turn out to be a important precedence. The European Union has taken important steps on this course with the introduction of the EU AI Act.
This weblog explores how the AI Act addresses safety for AI techniques and fashions, the significance of AI literacy amongst staff, and Cisco’s strategy for safeguarding AI by way of a holistic AI Protection imaginative and prescient.
The EU AI Act: A Framework for Safe AI
The EU AI Act represents a landmark effort by the EU to create a structured strategy to AI governance. One in every of its elements is its emphasis on cybersecurity necessities for high-risk AI techniques. This contains mandating sturdy safety protocols to forestall unauthorized entry and misuse, making certain that AI techniques function safely and predictably.
The Act promotes human oversight, recognizing that whereas AI can drive efficiencies, human judgment stays indispensable in stopping and mitigating dangers. It additionally acknowledges the essential position of all staff in making certain safety, requiring each suppliers and deployers to take measures to make sure a ample degree of AI literacy of their employees.
Figuring out and clarifying roles and tasks in securing AI techniques is complicated. The AI Act major focus is on the builders of AI techniques and sure normal goal AI mannequin suppliers, though it rightly acknowledges the shared accountability between builders and deployers, underscoring the complicated nature of the AI worth chain.
Cisco’s Imaginative and prescient for Securing AI
In response to the rising want for AI safety, Cisco has envisioned a complete strategy to defending the event, deployment and use of AI purposes. This imaginative and prescient builds on 5 key elements of AI safety, from securing entry to AI purposes, to detecting dangers similar to information leakage and complex adversarial threats, all the way in which to coaching staff.
“When embracing AI, organizations shouldn’t have to decide on between pace and security. In a dynamic panorama the place competitors is fierce, successfully securing expertise all through their lifecycle and with out tradeoffs is how Cisco reimages safety for the age of AI.”
- Automated Vulnerability Evaluation: By utilizing AI-driven methods, organizations can mechanically and constantly assess AI fashions and purposes for vulnerabilities. This helps establish a whole lot of potential security and safety dangers, empowering safety groups to proactively handle them.
- Runtime Safety: Implementing protections through the operation of AI techniques helps defend in opposition to evolving threats like denial of service, and delicate information leakage, and ensures these techniques run safely.
- Consumer Protections and Knowledge Loss Prevention: Organizations want instruments that forestall information loss and monitor unsafe behaviors. Corporations want to make sure AI purposes are utilized in compliance with inside insurance policies and regulatory necessities.
- Managing Shadow AI: It’s essential to watch and management unauthorized AI purposes, referred to as shadow AI. Figuring out third-party apps utilized by staff helps corporations implement insurance policies to limit entry to unauthorized instruments, defending confidential info and making certain compliance.
- Residents and staff coaching: Alongside the correct technological options, AI literacy amongst staff is essential for the secure and efficient use of AI. Growing AI literacy helps construct a workforce able to responsibly managing AI instruments, understanding their limitations, and recognizing potential dangers. This, in flip, helps organizations adjust to regulatory necessities and fosters a tradition of AI safety and moral consciousness.
“The EU AI Act underscores the significance of equipping staff with extra than simply technical information. It’s about implementing a holistic strategy to AI literacy that additionally covers safety and moral issues. This helps be sure that customers are higher ready to soundly deal with AI and to harness the potential of this revolutionary expertise.”
This imaginative and prescient is embedded in Cisco’s new expertise answer “AI Protection”. Within the multifaceted quest to safe AI applied sciences, rules just like the EU AI Act, alongside coaching for residents and staff, and improvements like Cisco’s AI Protection all play an essential position.
As AI continues to rework each trade, these efforts are important to making sure that AI is used safely, ethically, and responsibly, in the end safeguarding each organizations and customers within the digital age.
[1] Cisco’s 2024 AI Readiness Index
Share:
