Not If, But How: Responding to Security Incidents in the Age of AI Attackers

What happened
Over the past six months, attackers using AI tools have increased the frequency and speed of cyberattacks worldwide, targeting various layers of AI infrastructure and software supply chains. Security incidents were reported across package registries, SaaS platforms, self-hosted AI gateways, evaluation environments, and developer platforms. Companies such as Fireworks AI and Klue experienced breaches involving customer data and credentials. Security experts have outlined a four-part response strategy to address these rapid and complex threats.
Key details
- Between May and November 2026, at least 16 AI developer and hosting platforms disclosed security incidents.
- A campaign in May altered about 170 npm packages and some PyPI libraries, affecting SDKs from Mistral AI and Guardrails AI.
- In June, Klue reported misuse of OAuth tokens, exposing customer data from Salesforce and Gong.
- On November 16, code injected via Optimizely’s signup flaw reached 178 organizations’ websites, including HSBC and Salesforce.
- CrowdStrike observed one group using AI agents to attack over 200 organizations in 11 days.
Why it matters
Organizations using AI platforms, as well as their customers, now face a higher risk of rapid and large-scale breaches. Security teams must adopt automated detection, secure communication, thorough cleanup, and data minimization to limit damage and respond quickly. Companies that do not update their incident response processes may experience greater data loss and operational disruption as AI-driven attacks accelerate.
