How does SouthState govern and test its AI?
SouthState governs AI mainly through testing and a short list of approved tools. Only two generative-AI applications were approved for employees as of January 2025, Microsoft Copilot and OpenAI's ChatGPT, after vendor due diligence and tests written by subject-matter experts: half on accuracy and hallucinations, half on privacy, security, toxicity and jailbreaking, rerun for each new use case and as often as monthly. The practice dates to its first build: before Tate went live in May 2023 the bank 'red-teamed' the idea and had testers rate about 1,600 representative questions with a human-in-the-loop method, and built in citations and the model's 'thought process'. In 2026 the bank set three foundational objectives for its formal AI program, including standardising the AI risk framework. The 2024 and 2025 10-Ks list AI used by the bank and its vendors as a risk of errors, omissions, unfair treatment or fraud. Directors took AI-related continuing education in 2025, and a former Southern Company CIO chairs the board Risk Committee. SouthState has not published AI principles or a named AI ethics body.[1][2][3][4][5][6]
Data as of Oct 16, 2026 · SouthState profile updated Oct 13, 2026 · 18 sources cited on this page · 140 banks profiled on this site · Markdown
What AI controls has SouthState described?
Each control as the bank or its executives have described it, with the date of the source.
| Control | What it is | Date |
|---|---|---|
| Red-teaming before launch | Tate's use cases were 'red-teamed' to list constraints and risks; the bank chose an internal-only first use case because of 'the many unknowns in generative AI' | May 2023[2] |
| Human-in-the-loop accuracy test | A representative sample of about 1,600 questions rated by testers for accuracy, with the model's 'thought process', supporting content and citations shown for each answer | May 2023[2] |
| Prompt 'shots' as guardrails | Instructions placed before every question, such as not answering beyond the source documents and saying 'I don't know' when the documents fall short | May 2023[2][7] |
| Approved-tool list | Copilot and ChatGPT the only two approved gen-AI applications for employees; some public gen-AI uses stay off limits | January and March 2025[1][8] |
| Per-use-case model tests | Subject-matter experts write questions and expected answers; half test accuracy and hallucination, half privacy, security, toxicity and jailbreaking; the same model is retested for each new use | January 2025[1] |
| AI risk framework | 'Standardize AI risk framework' is one of three foundational objectives of the formal AI program | April 2026[3] |
| Agent controls (Fiserv agentOS) | The platform the bank is co-developing on includes identity-bound execution, policy enforcement and audit controls | May 2026[9][10] |
| Board oversight | Directors received AI-related continuing education in 2025; Martin Davis, a former Southern Company CIO, chairs the Risk Committee, which receives quarterly reports from the CIO and CISO | March 2026[6] |
How do SouthState's SEC filings describe AI risk?
The 2024 10-K, filed February 21, 2025, says the adoption and use of AI tools by the bank and its third-party vendors may increase the risk of errors, omissions, unfair treatment or fraudulent behaviour, and that generative AI 'has been known to, and may continue to, create biased, incomplete, inaccurate, misleading or poor-quality output'. It also notes that fraud attempts may accelerate as AI is used to produce sophisticated malware and phishing.[4]
The 2025 10-K, filed February 20, 2026, keeps that risk factor and adds the formal AI Strategy, tied to 'appropriate controls in place to maintain accuracy and reduce or eliminate bias'. The second-quarter 2026 10-Q repeats the AI risk language in its forward-looking statements.[5][11]
What has SouthState published on AI risk?
The SouthState Correspondent Division offers community banks an 'AI Risk Series', a short audio course on AI and its risks in banking that it says was 'adapted from our own internal training content here at SouthState'.[12]
Its 2023 posts argued that each generative model has a different risk profile and that model-risk groups should inventory the models touching the bank and rate them on accuracy, bias, security, transparency, data privacy, audit approach and ethics. Chris Nichols said in 2024 that answers varying from run to run is what makes gen-AI risk hard for bankers to assess.[2][13]
What does the SouthState profile record on this topic?
The rows below are the entries of the SouthState AI strategy profile that bear on this page, as of Oct 13, 2026; the profile holds the full record of 30 dated moves, 20 use cases and 33 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| Per-use-case model testing | Expert Q&A sets for accuracy, privacy, security, toxicity and jailbreaks; retested as models change. | In production | [1] |
| Co-developed agents on Fiserv agentOS | Agents within a controlled architecture with governance, audit trails and kill switches. | Announced | [10] |
| GenAI Knowledge Assistant | A chat interface for policy and procedure questions, launched in early 2025 and answering more than 1,100 questions a day. | In production | [14] |
| Call-centre AI | Real-time reference suggestions to agents during live calls, call sentiment monitoring, quality checks and wrap-up, and expanded self-service being tested. | Rolling out | [14][15] |
| Formal AI Strategy and AI risk-factor disclosure | The 10-K records a formal AI Strategy with controls to maintain accuracy and reduce or eliminate bias, and lists errors, unfair treatment and fraud from AI use as risks. | In production | [5] |
| Copilot Studio citizen development | The bank is treating Copilot Studio as a core capability so staff in business and risk functions can build agents under governance; the Q2 deck lists citizen agentic development under experimentation. | Pilot | [16][14] |
| AI embedded in vendor software | Management is deploying AI tools from its major software providers at department level and says many already embed AI in existing products; the 10-K expects vendors to add AI faster than the bank can, and its third-party risk team reviews third-party hosted applications holding sensitive data each year. | In production | [17][5] |
| AI risk framework and experimentation lab | Q1 2026 deck objectives: standardise the AI risk framework, educate all team members and launch an AI experimentation lab; the 10-K ties adoption to controls for accuracy and bias. | Rolling out | [3][5] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Jul 31, 2026 | Q2 2026 10-Q repeats the AI risk language | Forward-looking risks include technology change from AI and errors, omissions, unfair treatment or fraud from AI tools used by the bank and its vendors. | [11] |
| Apr 23, 2026 | Q1 2026 deck: a formal AI programme with three foundational objectives | Standardise the AI risk framework, educate all team members, launch an AI experimentation lab; internal knowledge search up 80% over twelve months. | [3] |
| Mar 3, 2026 | 2026 proxy: AI in director training; technologist chairs Risk Committee | Directors received AI-related continuing education in 2025; Martin Davis, a former Southern Company CIO, chairs the Risk Committee, which receives quarterly reports from the CIO and CISO. | [6] |
| Feb 20, 2026 | 2025 10-K: a formal AI Strategy, and AI as a risk factor | The bank says it has adopted a formal AI Strategy, plans to adopt AI 'internally and from vendors' with controls for accuracy and bias, and flags errors, unfair treatment and fraud risks from its own and vendors' AI. | [5] |
| Jun 30, 2025 | Independent core system conversion completed | The 10-K places completion of the core system conversion in the second quarter of 2025; no exact date is given. | [5] |
| Mar 20, 2025 | Gen-AI use policy in place of bans | Some public gen-AI uses remain off limits, per the bank's capital-markets head. | [8] |
| Jan 14, 2025 | Model testing before approval | Only Copilot and ChatGPT approved; expert-written accuracy and safety tests per use case. | [1] |
| Mar 18, 2024 | Nichols on the barrier to bank gen-AI use | In an American Banker survey story he says the difficulty of assessing the risk of a gen-AI application, because answers vary between runs, is what holds bankers back. | [13] |
| Jun 20, 2023 | Enterprise ChatGPT deployed on bank documents | Nichols, then director of capital markets, says the tool is trained only on bank documents, uses no customer data, and is expected to reach about 2,000 of 5,000 employees; he reports a five to eight times productivity gain on policy look-ups. | [18] |
How does SouthState test generative AI models?
Per use case. Chris Nichols told American Banker in January 2025 that subject-matter experts write questions and expected answers, half on accuracy and hallucinations and half on privacy, security, toxicity and jailbreaking, and that a model approved for one use must be retested before it is used for another. Tests are repeated as models change, as often as monthly.
Which AI tools may SouthState employees use?
As of January 2025, two generative-AI applications were approved for employees: Microsoft Copilot and OpenAI's ChatGPT, each after vendor due diligence and the bank's own testing. Some uses of public gen-AI tools remain off limits under the bank's policy. By mid-2026 Copilot had 2,500+ premium and 2,200+ active basic users.
Does SouthState disclose AI risks in its 10-K?
Yes. Both the 2024 and 2025 annual reports say AI use by the bank and its vendors may increase the risk of errors, omissions, unfair treatment or fraud, and that generative AI can produce biased, inaccurate or misleading output that may not be easily detected. The 2025 10-K adds a formal AI Strategy with controls for accuracy and bias.
Does SouthState have published AI principles?
No published AI principles document or named AI ethics board was found. What is on the record is the testing regime, the approved-tool list, a 2026 objective to standardise the AI risk framework, board AI training and the AI risk factor in its filings. Its correspondent division sells an AI risk course adapted from internal training.
Who oversees AI risk at SouthState's board?
The 2026 proxy says directors received AI-related continuing education in 2025 and that the Risk Committee, chaired by Martin Davis, a former Southern Company chief information officer and former Wells Fargo chief technology officer, receives quarterly reports from the chief information officer and chief information security officer.
- 6 things banks look for in gen AI models and companies
- Generative AI – 7 Lessons That Tate Taught Us
- SouthState first-quarter 2026 earnings presentation (Form 8-K, Exhibit 99.2)
- SouthState Corporation Annual Report on Form 10-K for the fiscal year ended December 31, 2024
- SouthState Bank Corporation Annual Report on Form 10-K for the fiscal year ended December 31, 2025
- SouthState Bank Corporation Definitive Proxy Statement (DEF 14A) for the 2026 annual meeting
- ChatGPT 3.5 Turbo – The Risks and Ethics for a Bank Enterprise Application
- Banks navigate workers' use of ChatGPT, set AI policies
- Fiserv launches agentOS operating system for banking AI agents
- Fiserv has co-created AI agents with six banks and OpenAI
- SouthState Bank Corporation Quarterly Report on Form 10-Q for the period ended June 30, 2026
- Introduction - AI Risk Series
- A third of banks ban employees from using gen AI. Here’s why
- SouthState second-quarter 2026 earnings presentation (Form 8-K, Exhibit 99.2)
- SouthState Bank Q2 Earnings Call Highlights
- The Playbook for Enterprise AI Adoption with Walt Wear (Community Bank Podcast)
- SouthState (SSB) Q1 2026 Earnings Transcript
- 'It's worth all the hype': SouthState Bank deploys enterprise ChatGPT
When SouthState moves on AI, you'll read it here first.
when one of the banks we cover moves on AI, it is in the next morning's brief · six sourced stories · 7 am ET · free
plus every tracker, bank and agent page update, the morning after · leave any morning