Salesforce's Trust Layer kept the high-stakes competition safe and secure.

Security Built Into Every Layer

Salesforce layered protection at every entry point of the experience — from the first click on the puzzle landing page all the way through every engagement with Slackbot — without adding a single point of friction for players. 

Salesforce Professional Services built this security framework from scratch, with no existing blueprint, anticipating threats and engineering safeguards that had never been needed on Salesforce before. 

At the perimeter:

• Akamai provided edge protection, bot detection, and geo-filtering before traffic ever reached the platform — serving 90% of static content via Akamai NetStorage as static HTML to keep it completely off the Salesforce backend.
• A dedicated AWS layer added another line of defense, with an API Gateway and web application firewall rate-limiting traffic to protect against DDoS attacks.
• Serverless Lambda functions routed users across 78 independent Salesforce shards so that if one environment went down, the others kept running.
• MuleSoft connected all 78 shards — along with every external platform and security tool in the stack — into a single, cohesive architecture, ensuring secure data flow across the entire experience.

Inside the experience:

• Toxicity detection scanned every user input and Slackbot response simultaneously to keep the experience appropriate for an all-ages audience.
• A machine learning-based prompt injection classification model ran in real time, layered with instruction defense, post-prompting review, and templated boundaries.

High-Volume Data Protection

To keep player data private, Slackbot only ever saw the bare minimum of information. Every conversation was processed in real time, but nothing was stored by external AI providers — and no player data was used to train third-party LLMs. Slackbot could only access what the logged-in user was already allowed to see.

A Redis cache held player profiles for 15 minutes at a time, so Slackbot could get most information quickly from memory instead of fetching it from Data 360 every time. That meant raw data was rarely retrieved directly, reducing the chances of interception or misuse. By design, this setup minimized exposure and kept sensitive player information protected at every step.

Security That Scaled With The Surge

When the Super Bowl commercial aired and millions of fans hit the platform at once, the architecture absorbed the surge without missing a beat. 

An Akamai waiting room acted as a pressure valve, letting users in at a controlled rate of 85,000 per minute to prevent database overload. Behind the scenes, DynamoDB partition sharding across more than 100 partitions gave millions of simultaneous writes a place to go, while AWS Private Connect kept all data transfers between the AWS layer and Salesforce off public networks.

“What made it possible was bringing together expertise across UI and UX, security and safety, scalability, and deep collaboration with engineering,” said Dressler. “We had the autonomy to build something we knew would work — something that would protect us, scale effectively, and anticipate the risks no one even knew to ask about yet, because no one fully understood what was possible.”

Over the course of the hunt, the system defended against 64,000 highly sophisticated prompt injection attacks — attempts to trick Slackbot into revealing puzzle answers. It blocked 99.997% automatically, with the remaining 0.003% caught by human oversight. 

Across 78 Salesforce shards, tens of thousands of LLM requests per minute, and 1 trillion tokens consumed in just a few days, the puzzle ran seamlessly, player data stayed secure, and zero toxic outputs were generated.