AccuSights AI Workforce
AI help that follows your rules.
AI agents for calls, email, invoices and paperwork, trained and run by us, each with its own login, firm limits, a human approver and a kill switch.
Try the AI Office
Every agent, every task, every approval.
This is the screen you get. Pick a sample office to see a morning. Try approving something, editing a draft, or pausing an agent. Nothing risky happens until you tap Approve.
- Working
- 2
- Waiting for you
- 2
- Tasks done today
- 80
- Guardrail events today
- 1
- Cost today
- $2.10
- ReceptionistActs within call scriptsWorking
Answering a call about a cleaning
Done today 38 - SchedulingActs within booking rulesWorking
Filling a 2 p.m. gap from the waitlist
Done today 14 - ReviewsNegative replies wait for youWaiting for you
Reply to a 2-star review needs your approval
Done today 6 - EvidenceRead onlyIdle
HIPAA evidence collected at 6 a.m.
Done today 22
Waiting for you
- Reviews
Reply to a 2-star review about wait times
Negative replies always wait for you. - Scheduling
Book an emergency visit at 4:30 p.m., after closing
Outside your booking rules.
Sample data. None of these offices is a client of ours; the names, counts and cost figures are illustrations of what the screen shows. Your office shows your own agents, approvals and costs.
Why now
Don’t get left behind. Don’t get burned.
Small businesses are already handing work to AI. Very few trust it to act alone yet, and most of the risk sits in how it is set up. That gap is where we work: start with one agent, prove it in 30 days, then add the next.
How we keep agents safe
Trained like staff. Watched like a vault.
We are a cybersecurity company first. Every agent starts at the lowest level that does the job, and only moves up when you say so.
- Level 0Read only
Looks and reports. Changes nothing.
- Level 1Draft
Prepares the work. A named person sends or saves it.
- Level 2Act with approval
Does the task after a person taps Approve in the AI Office.
- Level 3Act within limits
Acts alone inside hard limits (amount, recipients, hours, data types). Everything is logged and can be undone or paused.
- Its own login
- Every agent gets its own account with only the access its job needs. No shared passwords, no admin rights.
- A person approves the risky steps
- Money, deletions, new recipients and anything legal or medical wait for someone to tap Approve.
- Data boundaries
- Agents see only the folders, mailboxes and records we scope. Patient, defense and card data stay inside approved zones.
- Hidden instructions ignored
- Emails, web pages and files are treated as untrusted. Instructions hidden inside them are flagged and ignored.
- A full action log
- Every action is recorded: who did what, when and why.
- A kill switch
- One tap pauses one agent or the whole office.
- We try to trick it first
- Red-team tests before go-live and after every change, so we find the weak spot before a customer does.
- A budget
- Spend and rate limits on every agent. Runaway loops stop themselves.
- Watched by Protect
- If you have Protect, the agents’ accounts are connected during onboarding and watched like any employee account, with odd behavior flagged.
- A monthly review with you
- Quality, errors, blocked actions and cost, every month. Then we decide together what changes.
Aegis and Protect
The security office that can watch your agents too.
Aegis is the team of AI security agents inside AccuSights Protect, our AI-powered, human-enabled security service. If you have Protect, your agents’ accounts are connected during onboarding and watched like any employee’s, and an AccuSights engineer approves anything high risk.
The agent catalog
25 roles in 7 groups.
Open any role to see what it does, the limits it cannot cross and an example of its work. Security and Compliance comes first, because that is who we are.
Security and Compliance
Agents that keep you safe, audit-ready and honest about AI use.
Aegis SOC OfficeSix AI security agents and an Orchestrator that work every alert as a case, day and night, inside ProtectL3 for containment playbooks agreed at onboarding
What it does
- Mismatch Enrichment adds threat intelligence to every alert: the IP, the sender, the file, the device
- Dash SOC Triage decides whether the alert is real, how bad it is and what it touched
- AccuSights AI Analyst states the compliance impact: which HIPAA, CMMC or SOC 2 requirement the incident touches
- Security Monitoring Reporter writes the plain-English summary and the monthly monitoring report
- Policy Writer drafts the GRC policies your evidence has to match; Analyst Helper answers an analyst's questions about your logs, read only
- The Orchestrator routes each case through enrich, triage, compliance impact and reporting, then writes the case report
Limits it cannot cross
- Containment actions limited to an approved playbook list
- Anything high risk waits for an AccuSights engineer to approve it
- Every action reversible and logged, with the case report kept
- Cannot read mailbox or file contents beyond what an alert requires
When it works
24/7, every alert, in real time
Connects to
Endpoint protection (EDR), Microsoft 365 or Google Workspace, Firewall and VPN logs, Cloud accounts (AWS, Azure, Google Cloud), Identity provider
Level: L3 for containment playbooks agreed at onboarding; L2 for anything else
Why it pays: 31% of breaches start with vulnerability exploitation, ahead of stolen credentials for the first time in 19 years; AI is shrinking the time to exploit from months to hours. (Verizon 2026 Data Breach Investigations Report press release, 2026). Source
Comes with AccuSights Protect. Not sold separately.
EvidenceCollects proof that controls work: MFA reports, backup logs, training records, access reviewsL0 read only
What it does
- Collects proof that controls work: MFA reports, backup logs, training records, access reviews
- Maps each piece of proof to HIPAA, CMMC, SOC 2 or local regulator controls
- Flags missing or stale evidence weeks before an audit
- Builds the auditor folder and a one-page status for the owner
Limits it cannot cross
- Read-only API access
- No change to any system setting
- Evidence stored in client-owned storage with retention rules
- Assessor or AccuSights consultant reviews before anything is shared with an auditor
When it works
Daily collection, weekly summary, on demand before audits
Connects to
Microsoft 365 or Google Workspace admin, Backup console, Training platform, Ticketing system, AccuSights GRC workspace
Level: L0 read only
Why it pays: 63% of organizations studied have no AI governance policies to manage AI or prevent shadow AI. High levels of shadow AI added USD 670,000 to the average breach cost. (IBM Cost of a Data Breach Report 2025, 2025). Source
Vendor RiskKeeps a live list of every vendor that touches your dataL1 draft
What it does
- Keeps a live list of every vendor that touches your data
- Sends and chases security questionnaires and collects SOC 2 reports and BAAs
- Scores each vendor and flags expired contracts or missing agreements
- Watches news for breaches at your vendors and alerts you the same day
Limits it cannot cross
- Sends only to vendor contacts on an approved list
- Cannot sign or accept contracts
- Scores are advisory; a person decides
When it works
Weekly review, daily breach-news watch, at every new vendor
Connects to
Accounting system (to find who you pay), Email, Contract folder, AccuSights GRC workspace
Level: L1 draft; L2 to send questionnaires
Why it pays: Breaches involving a third party are up 60% and now account for 48% of all breaches. (Verizon 2026 Data Breach Investigations Report press release, 2026). Source
PolicyDrafts and updates security, privacy and acceptable AI use policies in your voiceL1 draft
What it does
- Drafts and updates security, privacy and acceptable AI use policies in your voice
- Answers staff questions about policy from your own approved documents
- Tracks who has read and signed each policy
- Flags when a law, framework or your own tools change and a policy needs a refresh
Limits it cannot cross
- Answers only from approved policy text and cites the section
- Every new or changed policy needs owner approval
- Never gives legal advice; routes legal questions to counsel
When it works
On request, monthly review, on regulatory change
Connects to
Document library, HR system, Microsoft Teams or Slack, AccuSights GRC workspace
Level: L1 draft
In the Aegis SOC Office inside Protect, this seat is called Policy Writer.
Why it pays: 13% of organizations reported breaches of AI models or applications; 97% of those lacked proper AI access controls. (IBM Cost of a Data Breach Report 2025 (Ponemon, 2025). Source
AI Use WatchFinds which AI tools staff are actually usingL0 read only
What it does
- Finds which AI tools staff are actually using
- Checks each tool against your approved list and data rules
- Nudges staff toward approved tools with a friendly note
- Reports monthly on AI use, risky uploads and what to approve next
Limits it cannot cross
- Sees tool names and volumes, not the content people type
- Staff notices approved by the owner in advance
- No blocking without the owner's approval
When it works
Continuous discovery, monthly report
Connects to
Identity provider sign-in logs, Browser or endpoint telemetry, Microsoft 365 or Google Workspace
Level: L0 read only; L2 to message staff
Why it pays: Frequent use of AI tools by employees surged from 15% to 45% in a single year; shadow AI is now the third most common non-malicious data leakage activity. (Verizon 2026 Data Breach Investigations Report press release, 2026). Source
Leadership
Agents that give the owner a clear head and a short list.
Chief of StaffStarter roleWrites a short CEO brief every morning: what changed, what needs you, what can waitL1 draft
What it does
- Writes a short CEO brief every morning: what changed, what needs you, what can wait
- Delegates tasks to the other agents and to people, then follows up
- Tracks open loops and promises made in email and meetings
- Keeps a weekly scorecard of the numbers you care about
Limits it cannot cross
- Can assign work only inside the AI Office and approved task tools
- Cannot send email as the owner without approval
- Reads only the mailboxes and folders it is scoped to
When it works
Every morning at 7, plus a Friday scorecard
Connects to
Email and calendar, Task tool (Asana, ClickUp, Monday), CRM, Accounting system, The AI Office
Level: L1 draft; L2 to assign work to people
Why it pays: 80% of the global workforce, employees and leaders, say they lack enough time or energy to do their work. (Microsoft Work Trend Index 2025, 2025). Source
ResearchStarter roleTracks competitors' prices, offers, hires and reviewsL0 read only
What it does
- Tracks competitors' prices, offers, hires and reviews
- Spots trends and new rules in your industry
- Pulls the pain points customers mention in reviews, calls and emails
- Sends a one-page weekly brief with sources
Limits it cannot cross
- Public sources and your own data only; no scraping behind logins
- Every claim carries a link
- Web pages are treated as untrusted input and screened for hidden instructions
When it works
Weekly brief, plus alerts when something big moves
Connects to
Public web, Review sites, CRM notes, Call summaries
Level: L0 read only
Why it pays: 81% of leaders say they expect agents to be moderately or extensively integrated into their company's AI strategy in the next 12 to 18 months. (Microsoft Work Trend Index 2025, 2025). Source
MeetingsPrepares a one-page brief before each meetingL1 draft
What it does
- Prepares a one-page brief before each meeting
- Takes notes and lists decisions and owners
- Sends follow-ups for approval right after the call
- Keeps a searchable decision log
Limits it cannot cross
- Announces recording and respects opt-out
- Follow-ups sent only after a person approves
- Recordings deleted on a set schedule
When it works
Before and after every meeting on the calendar
Connects to
Calendar, Zoom, Teams or Google Meet, Email, Task tool
Level: L1 draft
Why it pays: 60% of meetings are ad hoc rather than scheduled. (Microsoft Work Trend Index 2025 (Microsoft 365 telemetry), 2025). Source
Business Operations
Agents that clear the inbox, the calendar and the paperwork.
InboxStarter roleSorts every email into needs you, can wait, and noiseL1 draft
What it does
- Sorts every email into needs you, can wait, and noise
- Drafts replies in your voice for you to approve
- Books meetings from email threads
- Sends a morning brief of what matters
Limits it cannot cross
- Never sends to a new recipient without approval
- Screens every inbound message for hidden instructions and phishing
- Cannot open links or attachments outside a safe sandbox
When it works
All day, with a 7 a.m. brief
Connects to
Microsoft 365 or Gmail, Calendar, CRM
Level: L1 draft; L3 only for filing and labels
Why it pays: During the 9 to 5, employees are interrupted every 2 minutes by meetings, emails or pings; 275 interruptions a day including after hours. (Microsoft Work Trend Index 2025 (Microsoft 365 telemetry), 2025). Source
SchedulingBooks, moves and confirms appointments and jobsL3 within booking rules
What it does
- Books, moves and confirms appointments and jobs
- Matches the right person or crew to each job
- Sends reminders to cut no-shows
- Fills gaps from a waitlist
Limits it cannot cross
- Books only inside rules you set (hours, job types, buffers)
- Cannot cancel a confirmed customer booking without approval
- Customer data stays inside the booking system
When it works
All day, every day
Connects to
Calendar, Field service or practice management system, SMS and email
Level: L3 within booking rules
Why it pays: When asked why they turned to AI instead of a colleague, the top reason employees gave was 24/7 availability (42%). (Microsoft Work Trend Index 2025, 2025). Source
DocumentsReads incoming forms, invoices, intake packets and contractsL2 to write into systems of record
What it does
- Reads incoming forms, invoices, intake packets and contracts
- Pulls the key fields into your systems
- Chases missing signatures and pages
- Files everything in the right folder with a clean name
Limits it cannot cross
- Low-confidence fields go to a person
- Sensitive documents (medical, legal, CUI) only inside an approved data boundary
- No document leaves your storage
When it works
As documents arrive
Connects to
Email, Document storage (SharePoint, Google Drive, Dropbox), E-signature tool, Line-of-business system
Level: L2 to write into systems of record
Why it pays: Small business owners save a median 5 hours per week of their own time with AI tools, and businesses save a median 11.5 employee hours per week. (SBE Council, 2026). Source
IT HelpAnswers everyday IT questions for staffL1 draft
What it does
- Answers everyday IT questions for staff
- Walks people through password resets and device setup using approved steps
- Opens and routes tickets with the right details
- Spots repeat problems and reports them
Limits it cannot cross
- Verifies identity before any account help
- No admin rights; account changes need a person
- Answers only from the approved knowledge base
When it works
24/7 in Teams, Slack or email
Connects to
Microsoft Teams or Slack, Ticketing tool, Identity provider (read only)
Level: L1 draft; L2 for account actions
Why it pays: When asked why they turned to AI instead of a colleague, the top reason employees gave was 24/7 availability (42%). (Microsoft Work Trend Index 2025, 2025). Source
Finance
Agents that get you paid and show you the cash ahead.
FinanceStarter roleCategorizes expenses and matches receiptsL2
What it does
- Categorizes expenses and matches receipts
- Reconciles bank and card feeds
- Flags odd charges and duplicate bills
- Prepares a clean month-end pack for your accountant
Limits it cannot cross
- Cannot move money or pay bills
- Changes to closed periods blocked
- Every entry logged and reviewable by your accountant
When it works
Daily, with a month-end close
Connects to
QuickBooks or Xero, Bank and card feeds, Receipt capture, Email
Level: L2
Why it pays: Nearly 3 in 4 businesses (74%) aren't fully automated in bill management (QuickBooks wording: managing and paying bills). Manual processes are the number one internal reason for delayed outgoing payments. (Intuit QuickBooks, 2026). Source
CollectionsSends polite, well-timed invoice remindersL3 for standard reminders
What it does
- Sends polite, well-timed invoice reminders
- Adjusts tone by customer and how late the invoice is
- Offers easy payment links
- Escalates stuck invoices to you with a suggested next step
Limits it cannot cross
- Approved message templates only
- Never threatens, never offers discounts without approval
- Stops instantly when a customer disputes
When it works
Daily check, reminders on a schedule you set
Connects to
QuickBooks or Xero, Email and SMS, Payment links (Stripe, QuickBooks Payments)
Level: L3 for standard reminders; L2 for anything firmer
Why it pays: Nearly 3 in 5 small businesses (59%) have invoices overdue by 30+ days, up from 47% last year. Businesses with unpaid invoices are owed $17.7K on average. (Intuit QuickBooks, 2026). Source
Cash ForecastBuilds a rolling 13-week cash forecastL0 read only
What it does
- Builds a rolling 13-week cash forecast
- Shows payroll and big bills against expected receipts
- Warns early when a gap is coming
- Suggests which invoices to chase first
Limits it cannot cross
- Read only
- Forecasts labelled as estimates
- Finance data stays in your accounts and AccuSights' approved boundary
When it works
Weekly, plus alerts
Connects to
QuickBooks or Xero, Bank feeds, Payroll system
Level: L0 read only
Why it pays: Nearly 2 in 5 owners (39%) say one late payment made it hard to cover payroll or bills in the past year. (Intuit QuickBooks, 2026). Source
People
Agents that hire, onboard and answer staff questions.
HiringStarter roleWrites job posts from a short briefL1 for screening notes
What it does
- Writes job posts from a short brief
- Screens applicants against the must-haves you set
- Books interviews with candidates and your team
- Sends offer and onboarding paperwork for approval
Limits it cannot cross
- Never rejects a candidate on its own; a person makes every decision
- Screens only on job-related criteria you approve
- Bias and adverse-impact checks reviewed monthly
- Candidate data deleted on schedule
When it works
While a role is open
Connects to
Job boards, Applicant tracking system, Calendar, Email
Level: L1 for screening notes; L2 to contact candidates
Why it pays: Average nonexecutive cost per hire is $5,475; screening and interviewing alone average 8 to 9 days each. (SHRM press release, 2025). Source
OnboardingSends the welcome pack and paperwork before day oneL2
What it does
- Sends the welcome pack and paperwork before day one
- Requests accounts and equipment from IT
- Assigns training, including security awareness
- Checks in at day 7, 30 and 90 and reports back
Limits it cannot cross
- Account requests go through IT approval
- Personal data limited to what onboarding needs
- Check-in answers visible only to the manager and HR
When it works
From signed offer to day 90
Connects to
HR system, IT Help Agent, Training platform, Email
Level: L2
Why it pays: 47% of leaders list upskilling existing employees as a top workforce strategy for the next 12 to 18 months. (Microsoft Work Trend Index 2025, 2025). Source
HR AnswersAnswers staff questions about leave, benefits and the handbookL1
What it does
- Answers staff questions about leave, benefits and the handbook
- Fills simple HR forms
- Routes sensitive topics to a person at once
- Reports the most asked questions so you can fix the handbook
Limits it cannot cross
- Answers only from approved documents and cites them
- Complaints, medical or legal topics go straight to a person
- No access to pay or performance records
When it works
24/7 in Teams, Slack or email
Connects to
Handbook and policy library, HR system (read only), Teams or Slack
Level: L1
Why it pays: When asked why they turned to AI instead of a colleague, the top reason employees gave was 24/7 availability (42%). (Microsoft Work Trend Index 2025, 2025). Source
Sales and Marketing
Agents that answer leads fast and keep you visible.
SalesStarter roleReplies to new leads in seconds on web, email, SMS and chatL3 inside approved scripts
What it does
- Replies to new leads in seconds on web, email, SMS and chat
- Asks the qualifying questions you set
- Books the call on your calendar
- Follows up until the lead says yes, no or not now
Limits it cannot cross
- No prices, discounts or promises outside the approved sheet
- Honors opt-outs and contact rules (TCPA, CAN-SPAM)
- Hands hot or unhappy leads to a person
When it works
24/7
Connects to
Website forms and chat, GoHighLevel or your CRM, Calendar, Email and SMS
Level: L3 inside approved scripts
Why it pays: Only 31% of missed business calls get a callback at all, and nearly all of those happen within 24 hours. (Quo (formerly OpenPhone), 2026). Source
MarketingStarter roleDrafts social posts in your brand voiceL1 draft
What it does
- Drafts social posts in your brand voice
- Cuts short video clips from your longer videos
- Schedules posts across channels
- Reports what worked each week
Limits it cannot cross
- Nothing posts without approval until you raise the limit
- Brand and claims checklist on every post
- No client names or results without written consent
When it works
Weekly plan, daily posting
Connects to
Facebook, Instagram, LinkedIn, Google Business Profile, Video library, Scheduling tool or GHL social planner, Analytics
Level: L1 draft; L2 to publish
Why it pays: Leaders at Frontier Firms are far more likely to use AI for marketing tasks (73% vs 55% globally) and customer success (66% vs 44%). (Microsoft Work Trend Index 2025, 2025). Source
ProspectingBuilds lists of companies that fit your best customersL1 draft
What it does
- Builds lists of companies that fit your best customers
- Finds the right decision maker and a reason to reach out now
- Drafts short personal first emails
- Keeps the CRM clean and up to date
Limits it cannot cross
- Licensed data sources only
- Sending limits protect your domain
- Do-not-contact list always applied
When it works
Weekly list, daily drafts
Connects to
CRM, Data providers you license, Email
Level: L1 draft; L2 to send
Why it pays: Two-thirds (66%) of small businesses report revenue increases linked to AI; 22% report gains above 10%. (SBE Council, 2026). Source
ProposalsTurns call notes into a draft proposal or quoteL1 draft
What it does
- Turns call notes into a draft proposal or quote
- Pulls pricing from your approved price sheet
- Answers routine bid and security questionnaires from past answers
- Tracks opens and nudges you to follow up
Limits it cannot cross
- Prices only from the approved sheet
- A person signs off every proposal
- Questionnaire answers marked for review when unsure
When it works
On request
Connects to
CRM, Price sheet, Document templates, E-signature tool
Level: L1 draft
Why it pays: Small business owners save a median 5 hours per week of their own time with AI tools, and businesses save a median 11.5 employee hours per week. (SBE Council, 2026). Source
Customer
Agents that answer, book, help and thank customers on every channel.
SupportStarter roleAnswers FAQs on chat, email, SMS and socialL3 for answers from the knowledge base
What it does
- Answers FAQs on chat, email, SMS and social
- Handles simple requests such as order status, rescheduling and documents
- Works in every channel you use
- Escalates anything unusual to a person with the full story
Limits it cannot cross
- Answers only from approved knowledge
- Refunds and account changes need approval
- Always offers a human
When it works
24/7
Connects to
Help desk or GHL conversations, Knowledge base, Order or booking system
Level: L3 for answers from the knowledge base; L2 for refunds or account changes
Why it pays: By 2029, agentic AI will autonomously resolve 80% of common customer service issues without human intervention, leading to a 30% reduction in operational costs (prediction). (Gartner press release, 2025). Source
ReceptionistStarter roleAnswers every call, day and night, in a natural voiceL3 inside call scripts
What it does
- Answers every call, day and night, in a natural voice
- Books appointments straight into your calendar
- Routes urgent calls to the right person at once
- Logs every call with a short summary
Limits it cannot cross
- Says it is an AI assistant when asked and where the law requires
- Emergency words route to a person or 911 instructions at once
- No payment card numbers taken by voice
- Call recordings follow consent rules by state
When it works
24/7 or after hours only
Connects to
Phone system, Calendar or practice system, CRM
Level: L3 inside call scripts
Why it pays: 69% of missed business calls don't get a callback within 48 hours. (Quo (formerly OpenPhone), 2026). Source
ReviewsStarter roleAsks happy customers for a review at the right momentL2 for negative replies
What it does
- Asks happy customers for a review at the right moment
- Drafts a reply to every review
- Flags unhappy reviews to you within the hour
- Sends monthly themes: what people love and what to fix
Limits it cannot cross
- No incentives for reviews and no review gating
- Healthcare replies never confirm someone is a patient
- Negative replies always approved by a person
When it works
After every job or visit, daily replies, monthly themes
Connects to
Google Business Profile, Yelp, Facebook, CRM or booking system, SMS and email
Level: L2 for negative replies; L3 for positive replies inside templates
Why it pays: 89% of consumers expect business owners to respond to reviews. (BrightLocal, 2026). Source
Four weeks, then every month
Live in about four weeks, with you holding the approve button
- Week 1
We write each agent’s job description: tasks, limits, approvals, the data it may see.
- Week 2
We build it with its own login and limits, and connect your tools.
- Week 3
It drafts while your team does the work, and we try to trick it.
- Week 4
It goes live. Your staff learn to approve and pause in ten minutes.
- Every month
We watch it daily and review it with you monthly: hours saved, errors, cost.
Priced like staff, not like software
A one-time setup fee and a monthly fee, usage included up to a fair-use cap. You see the full number in writing before we build anything. Protect clients take 15% off every plan.
Pilot: One agent, 30 days, no card
A scoping call first, then one agent built, trained and run for 30 days with its own login, limits and an approver. Keep it, change it or stop; nothing renews on its own.
Starter: One or two agents
Design, build, training, red-team test, go-live and the monthly review. Usage included up to a fair-use cap.
Growth: Three to five agents
Everything in Starter for a small office of agents that hand work to each other, with one approver queue and one monthly review.
Full Office: Six or more agents
Every starter role you need, built and run by us, with the AI Office board and a named consultant.
- Protect clients take 15% off setup and monthly fees on every plan.
- Usage is included up to a fair-use cap on every plan; the cap is in the proposal.
- AI Workforce is provided by AccuSights LLC, in the United States.
- For scale: the average cost to hire one non-executive employee is $5,475, before salary (SHRM, 2025).
The Fit Check
Start with one agent
Five short steps, about three minutes. At the end you can start the $750 pilot (one agent, 30 days, no card, scoping call first), book a 30-minute demo, or ask for a custom proposal.
- A consultant reads every answer. No bots reply to this form.
- Patient data? We only use AI providers that sign a Business Associate Agreement.
- Defense data stays out of scope unless it runs inside your approved enclave.
Questions about AI agents
Will AI agents replace my staff?
Is it safe to let an agent near patient or client data?
How do I stop an agent?
What does AI Workforce cost?
Is Aegis part of AI Workforce?
Standards we build and test against
- OWASP Top 10 for LLM Applications prompt injection, sensitive data disclosure, excessive agency and the rest
- OWASP Top 10 for Agentic Applications (December 2025): goal hijack, tool misuse, identity and privilege abuse, rogue agents
- NIST AI Risk Management Framework 1.0 and its Generative AI Profile (NIST AI 600-1): the backbone of each agent’s charter and the monthly review
Which control stops which risk
- Own login, least privilege: identity and privilege abuse, excessive agency
- Approval gates: tool misuse, people trusting an agent too much
- Input screening: prompt injection and goal hijack
- Action log and kill switch: cascading failures and rogue agents
- Pinned, vetted models and connectors: supply chain risk
Compliance angles
- HIPAA: we sign a BAA, use only providers that sign one for the exact service, and never use consumer AI accounts with patient data. Healthcare and dental
- CMMC: agents do not touch CUI by default; when they must, they run inside your enclave. The evidence agent is read only and certifies nothing. CMMC readiness
- SOC 2: agent identities, monitoring, change control and red-team results drop into your evidence. SOC 2 readiness
- Payment cards: no agent takes card numbers by voice or chat. Payment links only.
AI Workforce is a managed service of AccuSights LLC. Agent roles, limits and examples on this page are described as designed; example outputs are sample data.
Next step
Start with one agent. Keep the approve button.
Tell us which jobs eat your week. Start the $750 pilot, see the AI Office in a 30-minute demo, or get a written proposal within one business day.