DARPA selects 17 orgs for its AI defense program, with Intel leading efforts to defend against physical attacks like use of stickers to trick self-driving cars
think facial recognition programs, voice recognition tools, self-driving cars, weapon-detection software, etc. https://www.protocol.com/...
Context & Ripple Effects
DARPA has been running this playbook for years: the Cyber Grand Challenge pitted AI hacking systems against each other in 2016, and the $2B 'AI Next' investment campaign followed in 2018. The 2023-24 AIxCC contest with Anthropic, Google, Microsoft, and OpenAI extended the template to scanning open-source code for flaws.
The new program applies that same contest structure to a different attack surface: the physical world. Instead of code vulnerabilities, the 17 selected organizations — with Intel leading the self-driving-car effort — are tasked with defending deployed AI like facial recognition, voice recognition, and weapon-detection software against adversarial inputs such as stickers that trick perception systems.
First-order effects
- Intel now leads a DARPA-backed team hardening self-driving-car perception against physical adversarial attacks, giving it a marquee government validation for its AI work beyond silicon.
- Sixteen other organizations receive DARPA's imprimatur and resources to defend facial recognition, voice recognition, and weapon-detection systems — a direct funding and credibility boost for each.
Second-order effects
- Autonomous-vehicle makers outside the program now face a government-endorsed benchmark for adversarial robustness, pressuring them to fund comparable defenses or explain why their perception stacks lack them.
- Rival chip and AI vendors not among the 17 must weigh whether to seek similar defense-program roles, since DARPA selection is becoming a credential that shapes who gets considered for follow-on government AI work.
Third-order effects
- If the pattern holds — Cyber Grand Challenge to AI Next to AIxCC to physical-attack defense — DARPA's competition pipeline is becoming the standard mechanism for hardening AI before deployment, moving adversarial robustness from an academic concern to a de facto procurement requirement.
- The program signals that defense against AI manipulation is splitting into two tracks, digital (AIxCC's code scanning) and physical (Intel's sticker defense), and vendors able to cover both will be structurally advantaged in national-security AI contracts.
The trend: DARPA is extending its contest-driven AI-hardening pipeline from software vulnerabilities to physical-world attacks, with Intel positioning itself as a defense AI prime in the process.