Most companies rolling out a customer-service AI have a forgiving failure mode: worst case, it gives a bad answer and someone calls in annoyed. TSA doesn’t get that luxury. Get the answer wrong on what counts as a “liquid” or how a CPAP machine gets screened, and someone’s repacking a bag at the checkpoint, or standing there without something they could’ve prepared for the night before. That’s the environment TSA rolled its new AI agent into anyway.

Ace is already doing more volume than most customer service teams see in a year. Salesforce announced the deployment September 14, 2026. Ace runs on Agentforce, Salesforce’s agentic AI product, built specifically for TSA through Salesforce Public Sector Solutions and secured with Salesforce Shield. It’s live now on tsa.gov and inside the myTSA app, and it’s been quietly fielding traveler questions since sometime this summer, ahead of the holiday travel crush everyone in the industry is already bracing for.

Here’s the number that actually matters: roughly 100,000 traveler conversations a month, and Ace resolves 96% of them without ever routing to a human. That’s not a pilot running in a lab somewhere. That’s a production system already carrying real checkpoint-anxiety traffic at scale.

Quick Verdict

What it is	An AI agent, built on Salesforce Agentforce, that answers traveler questions on tsa.gov and myTSA
Announced	September 14, 2026
Volume	~100,000 traveler conversations per month
Resolution rate	96% resolved without a human agent
Trained on	TSA’s federal website content plus 550+ knowledge articles
Where to find it	tsa.gov and the myTSA app
What it’s good for	Routine questions: liquids packing, medical device screening, checkpoint basics
What it’s not	A replacement for checking current rules on anything airport-specific or unusual

What Is TSA’s Ace AI Agent?

Ace is an AI-powered chat assistant, built by Salesforce for TSA, that answers common traveler questions about airport security on tsa.gov and the myTSA app. It’s trained on TSA’s own published content and a library of more than 550 knowledge articles, so it sticks to what TSA already publishes rather than reaching into the open internet for an answer. Routine questions (can I bring this liquid, how does a medical device get screened) get resolved directly. Anything Ace can’t confidently answer gets escalated to a live TSA staffer, along with an AI-generated summary of the conversation so the human doesn’t start from zero.

That escalation path isn’t just good customer-service design — it’s close to a compliance floor. Federal agencies deploying AI tools are bound by OMB Memorandum M-24-10, which requires minimum risk-management practices, including human review, before an agency can deploy AI that’s “safety-impacting.” A public chatbot fielding questions about airport security screening lands squarely in that category. Ace’s hard handoff to a live officer whenever it isn’t confident, wrapped in Salesforce Shield’s encryption and logging, reads less like a nice-to-have feature and more like the minimum TSA had to build to put a chatbot in front of checkpoint questions at all.

What Does Ace Actually Answer?

TSA and Salesforce point to a narrow, practical set of use cases, the stuff that generates the bulk of TSA’s contact center volume in the first place:

1. Liquids and 3-1-1 packing questions. What counts as a liquid, what fits in the quart bag, what needs a separate bin.
2. Medical device and checkpoint accommodation questions. How insulin pumps, CPAP machines, and similar devices get screened, and what to expect walking up with one.
3. General checkpoint procedure questions. What to take out of your bag, what’s allowed in a carry-on versus checked, basic ID requirements.
4. Simple status and policy lookups pulled straight from TSA’s own published guidance, the same info you’d find by digging through tsa.gov yourself, just faster.

That fourth category is the real value proposition. Most of what Ace answers isn’t complicated — it’s just buried three clicks deep on a federal website, or it’s the kind of question that used to sit in a phone queue behind actual emergencies.

The Numbers Behind the Rollout

Salesforce’s announcement leans hard on cost and scale, and the figures are specific enough to be worth repeating exactly:

• ~100,000 traveler conversations per month, live since Ace went into production during this year’s summer travel surge.
• 96% resolved without human escalation — meaning only about 1 in 25 conversations needs a person.
• Per-conversation cost cut by more than 90%, according to Salesforce, down from over $4 per interaction.
• 11,660+ staff hours saved annually, per TSA’s broader digital-modernization figures that Salesforce cites alongside Ace.

Worth separating out: the conversation volume and resolution rate are corroborated by Government Executive and Nextgov/FCW, both citing TSA directly rather than repeating a press release. The 90% cost cut isn’t independently verified the same way — that’s Salesforce’s own arithmetic about a product it sold to TSA.

But the savings figure isn’t the number worth sitting with here. The other side of 96% is: roughly 4,000 of every 100,000 monthly conversations that Ace can’t resolve and hands to a person. For a retail chatbot, that 4% just means someone waits on hold. For a security agency’s public-facing tool, TSA needs that handoff to work cleanly every time, especially during a holiday surge, because the questions that stump Ace are, almost by definition, the ones where a dead end or a wrong guess has consequences beyond an annoyed customer.

How Ace Fits Into TSA’s Bigger AI Push This Year

Ace didn’t show up in isolation. TSA has spent 2026 stacking automation and biometric tools on top of each other, mostly aimed at PreCheck members: Touchless ID’s facial-comparison rollout hit 65 airports earlier this year, Gateside let non-ticketed PreCheck members walk companions to the gate starting in September, and Newark’s AutoCAT eGates removed the officer from the ID-check step entirely, at least for one terminal. Ace is the front-end version of the same instinct. It automates the parts of the traveler experience that don’t require a person and saves the humans for the parts that do.

It also lands in the same month TSA has been weighing an exception to the 3-1-1 liquids rule for PreCheck members at CT-scanner lanes. None of that is confirmed yet, but if it does roll out, expect “can I bring a full water bottle through my airport” to become one of Ace’s most common questions within a week.

What Ace Can’t Do

Be clear-eyed about the boundaries here, because “96% resolved” doesn’t mean “96% of everything you might ask.”

Ace answers from TSA’s own published content, which means it’s only as current as what TSA has posted. It won’t tell you whether your specific airport has a CT scanner, or anything airport-by-airport that TSA hasn’t centrally documented (including whether a policy that’s still “considering” something has actually taken effect). The $45 TSA ConfirmID fee for REAL ID exceptions is exactly the kind of policy detail Ace should nail, since it’s a documented federal rule. Whether your local terminal happens to be running a pilot program is a different kind of question entirely, and one no chatbot trained on official web copy is going to answer better than checking the specific program’s own page.

The bigger caution, and one we’ve written about at length in our guide to verifying AI travel planner answers: any AI assistant can sound confident while being wrong, especially on edge cases that fall outside its training material. The escalate-when-unsure design is supposed to catch that failure mode before it reaches you, which is exactly the human-review safeguard federal risk-management rules call for. It’s a safeguard, not a guarantee. If Ace gives you an answer that determines whether you check a bag or leave something at home, and the stakes are high enough to matter, cross-check it against TSA.gov directly before you pack.

Should You Trust What Ace Tells You?

For the routine stuff (3-1-1 liquids, what’s allowed in a carry-on, how a specific medical device typically gets screened), yes, reasonably. That’s exactly the question set Ace is built and trained for, and it’s pulling from TSA’s own documentation. A 96% resolution rate across 100,000 monthly conversations means it’s already been tested against real traveler questions at a volume most customer service tools never see.

For anything airport-specific or tied to an undocumented pilot program, or anything where the answer changes what you pack or how you plan your arrival time, verify it against TSA.gov directly, or call ahead. The downside of a wrong answer on a routine liquids question is mild annoyance at the bin. Trusting a wrong answer on whether your airport runs a specific screening program means standing in the wrong line with the wrong expectations.

Who Should Actually Use This

Anyone with a quick, routine security question before a flight. This is squarely what Ace is for, and it’s faster than digging through TSA.gov’s own navigation or waiting on a phone line.

Travelers with medical devices who want to know what to expect. Screening procedures for insulin pumps, CPAP machines, and similar equipment are exactly the kind of documented, repeatable question Ace handles well.

Anyone flying during the holiday scale-up who wants an answer at 11 p.m. Ace doesn’t keep contact-center hours. That alone covers a real gap. TSA phone support isn’t staffed around the clock, and a lot of pre-trip anxiety happens the night before a flight.

Who Should Skip It

Anyone asking about a specific, still-unconfirmed policy change. If a program is still “under consideration” (like the possible PreCheck liquids exception), Ace’s answer is only as good as what TSA has already published, which for an unconfirmed policy is often “not much.”

Anyone with a genuinely unusual situation. Ace escalates what it can’t confidently answer, which is the right design. But if your question involves an edge case (an uncommon medical device, an unusual document situation, a dispute about how a checkpoint handled something), go straight to a human rather than looping through the chatbot first.

The Bottom Line

Ace is a real, working AI deployment already carrying meaningful traveler-support volume, not a demo dressed up as a launch, and outlets that talked to TSA directly back up the conversation count and resolution rate. What’s worth watching isn’t the savings figure Salesforce is promoting about its own product — it’s whether that 4% handoff to a person holds up cleanly once holiday volume hits, since a dropped handoff here means a traveler stuck at a checkpoint with no good next step, not just a missed discount. Use Ace for what it’s built for: routine, documented questions about packing and screening. For anything airport-specific, or riding on a policy that’s still “under consideration,” skip the chatbot and check TSA.gov directly.

---

Details current as of September 18, 2026, based on Salesforce’s September 14 announcement and reporting from Government Executive and Nextgov/FCW. Usage volume and resolution rates will shift as adoption grows. Verify current TSA policy at TSA.gov before you fly.