# Conditional Trust: The Public Blueprint for Acceptable AI **Author:** G2 Research **Published:** June 2026 **Based on:** This research draws on 92 in-depth interviews with business professionals representing a wide mix of roles, industries, and company sizes. **Topics covered:** AI Trust & Governance How 92 business professionals are navigating AI acceptance and trust: 84% take a cautious but practical orientation toward AI, 62% say training data is fair only with consent and compensation, 66% support AI in hiring only when human review is required, and 53% say managing AI-driven job disruption is a shared responsibility of employers and government. **Executive answer:** 84% caution, not rejection, defines ai attitudes: Most respondents take a pragmatic but cautious stance toward AI, showing that adoption barriers are less about hostility to the technology and more about whether its use feels legitimate and controlled. **On this page:** [Overall Value Orientation Toward AI](#overall-value-orientation-toward-ai) · [Privacy Risk Frame in AI Systems](#privacy-risk-frame-in-ai-systems) · [Transparency and Disclosure Expectations](#transparency-and-disclosure-expectations) · [Training Data Fairness Boundaries](#training-data-fairness-boundaries) · [Acceptance of AI in Hiring and Promotion](#acceptance-of-ai-in-hiring-and-promotion) · [Human Versus AI Decision Authority](#human-versus-ai-decision-authority) · [Responsibility for Managing AI Job Disruption](#responsibility-for-managing-ai-job-disruption) · [Adaptation and Upskilling Posture](#adaptation-and-upskilling-posture) · [Acceptance Conditions for AI Surveillance](#acceptance-conditions-for-ai-surveillance) · [Deepfake Acceptability and Ethical Boundaries](#deepfake-acceptability-and-ethical-boundaries) · [Primary Harm Expected from Deepfakes](#primary-harm-expected-from-deepfakes) · [Perceived Risks of AI-Mediated Hiring](#perceived-risks-of-ai-mediated-hiring) · [Outlook on AI-Driven Job Disruption](#outlook-on-ai-driven-job-disruption) · [Strategic takeaways](#strategic-takeaways) · [Frequently asked questions](#frequently-asked-questions) · [How this was produced](#how-this-was-produced) --- ## Overall Value Orientation Toward AI AI is viewed primarily as a practical tool rather than a transformational force. Roughly 84% of respondents who discussed it took a pragmatic but cautious stance, emphasizing usefulness alongside the need for oversight. Only 15% expressed clearly AI-positive enthusiasm, while just 2% focused mainly on harms. This pattern points to conditional adoption: respondents are open to efficiency gains and broader benefits, but resist unchecked deployment, especially when jobs, governance, and accountability are at stake. The small strongly positive segment highlights productivity and societal upside, while the tiny skeptical minority centers on privacy and surveillance risks. **Key takeaways:** - Pragmatism defines AI sentiment: 84% described their stance as pragmatic but cautious rather than purely enthusiastic or dismissive - Support is broad but rarely unqualified: 59% were pragmatically positive, while only 11% were strongly optimistic and 30% expressed mixed positivity - Human oversight is nonnegotiable: 100% accepted AI only with strong guardrails, making governance a universal condition of support **Strategic implication:** Lead with governed, human-in-the-loop AI offers, not autonomy-first positioning. Package core features with visible approval controls, audit trails, escalation paths, and role-based permissions, then price advanced automation as opt-in modules tied to compliance and risk management outcomes. Message AI as a productivity and decision-support tool that improves speed and consistency while preserving human accountability, and equip sales and implementation teams to prove guardrails before promising innovation. --- ## Privacy Risk Frame in AI Systems Privacy concerns around AI are led by fears of surveillance and hidden collection, cited by 46% of respondents. Another 36% focused on data misuse, storage, and profiling, showing that concern extends beyond being watched to how personal information is retained, connected, and commercially exploited. Respondents described a risk chain that starts with invisible collection and escalates into profiling, tracking, and secondary use without consent. Only a small minority framed privacy as a tradeoff, suggesting organizations need stronger limits on data capture, retention, and downstream use, especially where commercial incentives make monitoring feel exploitative. **Key takeaways:** - Surveillance fears dominate privacy risk: 51% describe localized or situational monitoring concerns, while 30% point to mass surveillance and systemic privacy erosion and 19% fear always-on or hidden monitoring - Data misuse rivals monitoring anxiety: 48% express strong concern about data storage, reuse, or profiling, nearly matching the share focused on surveillance and hidden collection overall at 46% - Control matters, but acceptance is limited: only 22% say privacy is manageable through consent, notice, or safeguards, while 27% accept privacy loss as a tradeoff for safety or convenience **Strategic implication:** Prioritize privacy-by-design in product and go-to-market: make data collection visible, minimize always-on sensing, default to local processing and short retention, and give users granular controls over monitoring, storage, and downstream reuse. Position premium trust through auditable safeguards, clear purpose limitation, and anti-profiling commitments rather than convenience alone. Segment offers by risk tolerance, reserving broader data-sharing features for explicit opt-in tiers and regulated, high-scrutiny contexts. --- ## Transparency and Disclosure Expectations Basic transparency is the baseline expectation: 46% of respondents emphasized simple disclosure that AI is being used, while 29% went further and rejected black-box systems outright. Another 25% focused on explainability in high-stakes decisions, showing that trust starts with visibility but often rises with the consequence of the outcome. Expectations become stricter as risk increases. In lower-stakes contexts, respondents mainly wanted upfront notice and clarity on data use; in hiring, credit, health care, or legal decisions, many expected traceable reasoning that humans could review. This creates a tiered mandate for organizations: disclose AI use by default, and provide stronger explanation where decisions materially affect people. **Key takeaways:** - Traceable disclosure is the trust baseline: 66% want disclosure plus source traceability, while only 15% accept basic notice alone and 13% expect full provenance and process visibility - Explainability is expected, especially when stakes rise: 42% want explanations mainly for important or high-stakes decisions, and 44% require explanations in all cases, leaving just 10% satisfied with limited or on-demand explanation - Black-box AI still faces a hard trust ceiling: with 46% emphasizing disclosure and trust-building transparency, expectations cluster around visibility and accountability rather than opaque AI experiences **Strategic implication:** Make traceable disclosure the default product standard: surface AI use clearly, cite sources, and provide decision-level explanations at minimum for high-stakes outcomes, with always-on explainability in regulated or trust-sensitive contexts. Package transparency as a tiered offer—baseline disclosure for low-risk workflows, deeper provenance and audit trails for enterprise and compliance buyers—and shift messaging from automation speed alone to verifiable accountability, human oversight, and reduced adoption risk, since opaque AI will cap trust and conversion. --- ## Training Data Fairness Boundaries Consent and compensation set the clearest boundary for fair AI training data use, with 62% saying both are required. By comparison, only one in five view public data as fair game, and 17% accept use under legal or conditional fair-use rules, making unrestricted scraping a minority position. Public availability alone does not resolve concerns for most respondents; the bigger issue is whether creators and users knowingly agreed and share in the value created. This creates a practical mandate for AI companies: transparent disclosure, explicit permission, and compensation models will resonate more strongly than relying on public-domain or ambiguous legal defenses. **Key takeaways:** - Consent and compensation set the fairness line: 68% say training data use is only fair when creators give consent and receive compensation or credit, and 62% overall define fairness through these requirements - Permission outweighs mere access: while 26% say publicly available data is fair game, only 19% rely on legality or copyright alone, versus 68% who require consent plus compensation or credit - Utility arguments remain secondary to creator rights: 20% justify broad data access for innovation or usefulness, but this trails far behind the 68% who prioritize consent and compensation and the 16% who specifically require consent for personal or creator data **Strategic implication:** Build training-data strategy around licensed, opt-in supply: secure explicit creator consent, attach compensation or credit by default, and make provenance and usage terms visible in product and policy messaging. Shift investment from “publicly available” scraping toward rights-managed partnerships, creator marketplaces, and auditable consent workflows. Price enterprise offerings to reflect compliant data sourcing, and position the premium around lower legal risk, stronger brand trust, and creator-aligned AI development. --- ## Acceptance of AI in Hiring and Promotion AI is acceptable to most candidates only at the front end of talent decisions. About two-thirds support AI for screening when humans still review outcomes, while 28% reject any AI role in hiring or promotion decisions. Just 5% are open to a broader AI role, and even they expect strong safeguards. Support centers on efficiency tasks such as summarizing resumes, sending reminders, and narrowing options, not deciding who gets hired or promoted. The clearest fault line is authority: most respondents want humans accountable for consequential decisions, while a smaller group also worries AI could strip out human judgment or amplify bias in high-stakes employment choices. **Key takeaways:** - AI is welcomed only as a filter: 42% accept AI for screening and support, and another 43% accept it only when human review is built in, showing broad support for AI in an assistive rather than autonomous role - Humans must make the final call: 78% say final authority must stay with people, while only 8% express conditional trust in AI with transparency and guardrails - Promotion decisions are where AI loses support: 13% strongly reject AI decision-making and 11% reject even support use, reinforcing that acceptance drops sharply when AI moves from screening into consequential career decisions **Strategic implication:** Position AI as a screening and workflow-acceleration layer, not a decision-maker, and build every hiring and promotion offering around explicit human sign-off. Package promotion-related features as decision support with audit trails, transparent criteria, and recruiter or manager review checkpoints, while avoiding autonomous recommendation claims in messaging. Price premium tiers around governance, explainability, and compliance controls, and market the product on speed, consistency, and reduced admin burden rather than replacing human judgment. --- ## Human Versus AI Decision Authority Leaders largely want AI to inform employment decisions without controlling them. Roughly half, 47%, support conditional delegation with human oversight, while 46% limit AI to a support role only. Just 7% trust AI to make some decisions independently, showing that final authority remains firmly with people. This split signals broad comfort with AI handling analysis, screening, and recommendation tasks, but not accountability for high impact outcomes. The distinction between conditional delegation and support-only use is narrow, suggesting most respondents accept AI in the workflow if humans review, validate, and own the final hiring or promotion call. **Key takeaways:** - Human authority remains the red line: 82% insist humans keep final say, with 35% requiring human-only authority and 47% allowing AI input only in lower-stakes or reviewed decisions - AI is a filter, not the decider: 74% support AI handling screening or first-pass work with human review, while just 4% are comfortable with AI doing most of the process under human veto - Conditional delegation defines the dominant model: 47% favor using AI as a conditional advisor under human oversight, and only 16% want AI limited to narrow support roles alone **Strategic implication:** Position AI as a triage and recommendation layer, not an autonomous decision-maker: build workflows where AI handles screening, prioritization, and first-pass analysis, while humans retain explicit approval rights, especially for high-stakes cases. Price and package around governed oversight features—review queues, audit trails, confidence flags, and veto controls—and lead messaging with “human in command” assurance. Target adoption by selling risk reduction, speed, and consistency without surrendering executive accountability. --- ## Responsibility for Managing AI Job Disruption Shared responsibility is the dominant view on managing AI job disruption. A majority, 53%, say employers and government should share the lead, while another 11% broaden that responsibility across employers, government, and individuals. By contrast, only 7% place the burden primarily on individual workers to adapt on their own. This points to a strong expectation that institutions, not just employees, must absorb the shock of AI-driven change. Respondents consistently pair employer duties such as ethical deployment and worker support with government roles in regulation, training, and social policy, while individuals are seen as contributors rather than the main line of defense. **Key takeaways:** - Shared accountability is the dominant expectation: 76% say employers and government share responsibility for managing AI job disruption, making this by far the leading view - Purely individual responsibility remains a minority stance: just 13% say individuals are primarily responsible, far below the 53% who explicitly frame the response as shared between employers and government - Adaptation is seen as collective, not solo: 30% say responsibility is shared but adaptation centers on individuals, while only 8% put primary responsibility on employers and 12% on government alone **Strategic implication:** Build AI workforce strategies as employer-led programs backed by public-sector partnerships, not employee self-service offerings alone. Package retraining, redeployment, and transition support as shared-accountability solutions, with employers funding implementation and government offsetting costs through grants, tax incentives, or workforce programs. Position messaging around joint responsibility and practical worker protection, and price offerings for enterprise and institutional buyers rather than individuals, reserving direct-to-worker products as a secondary layer. --- ## Adaptation and Upskilling Posture Proactive AI upskilling sets the tone, with 55% of respondents describing active efforts to adopt AI and build capability. Another 34% take a selective, bounded approach, using AI where it adds efficiency but keeping guardrails in place. Only 11% center their perspective on risks of overreliance. This creates a workforce posture that is more pragmatic than polarized. Most respondents are not resisting AI, but many are defining clear limits around decision-making, privacy, and human judgment. In practice, the leading pattern is experimentation paired with oversight, while a smaller group worries that excessive dependence could erode expertise and cognitive development. **Key takeaways:** - Proactive adaptation is the clear norm: 94% describe either pragmatic adaptation or proactive upskilling, while just 1% resist or avoid adapting to AI - Upskilling is as common as pragmatic use: 47% report proactive AI upskilling and career use, matching the 47% who see adaptation as necessary and practical - AI use stays firmly human-led: 87% favor selective use with human oversight, while only 13% center concern about harm to human thinking or agency **Strategic implication:** Build offerings around “human-in-the-loop” productivity gains: package AI as a practical co-pilot with clear oversight controls, auditability, and role-specific training pathways. Price and position in tiers that support both pragmatic adopters and proactive upskillers, pairing core workflow automation with premium enablement, certification, and career-oriented learning. Lead messaging with augmentation, safety, and accountability—not autonomy—while equipping managers to set usage boundaries, governance standards, and measurable skill-development goals. --- ## Acceptance Conditions for AI Surveillance Nearly two-thirds, 64%, accept AI surveillance only when it comes with clear notice, legal guardrails, and strict limits on use and retention. Far fewer take a more permissive view, with 15% broadly supporting AI surveillance for safety, while 18% reject it outright in all cases. This pattern points to conditional legitimacy rather than blanket approval. The largest group supports AI surveillance for specific security or monitoring purposes, but only when it is necessary, proportionate, and governed in advance. By contrast, broad supporters focus on crime reduction benefits, while opponents center on accountability risks and the danger of expanded enforcement. **Key takeaways:** - Conditional acceptance dominates views: 66% accept AI surveillance only under specific conditions, while just 17% support broad public-safety use and 17% reject it outright - Clear limits are the price of acceptance: 64% accept AI surveillance only with notice and limits, and 52% specifically restrict it to narrow security exceptions or exclude private spaces - Transparency matters more than blanket approval: 30% require notice, consent, or visible awareness, while another 14% say strict data-use protections and oversight are necessary **Strategic implication:** Lead with a “bounded use” offer: deploy AI surveillance only in clearly defined security scenarios, exclude private or sensitive spaces by default, and make visible notice, consent options, and retention rules part of the standard product. Package governance as a core feature—not an add-on—with audit trails, access controls, and independent oversight. Position and price solutions around compliance, transparency, and risk reduction rather than broad surveillance capability. --- ## Deepfake Acceptability and Ethical Boundaries Consumers largely reject deepfakes unless clear guardrails are in place. Nearly half, 46%, are outright opposed to deepfakes, while about one-third, 32%, would allow them only with explicit consent and disclosure. Just 23% accept them in narrow entertainment or satire contexts, showing broad discomfort with synthetic media. Acceptance depends on transparency and permission, not enthusiasm for the technology itself. Even among those open to limited use, support centers on watermarking, disclosure, and consent, while entertainment remains the main exception. In practice, brands and platforms face a high bar: unlabeled or unauthorized deepfakes risk immediate consumer backlash. **Key takeaways:** - Deepfakes face strong baseline rejection: 46% were outright opposed to deepfakes, including 38% who said they should be banned entirely - Most acceptance is tightly limited: 52% would allow deepfakes only in narrow, low-stakes situations, while just 10% support broader entertainment or creative use - Consent and disclosure define the ethical line: 72% said deepfakes are unacceptable when deceptive or harmful, and only 28% find them acceptable with clear consent and disclosure **Strategic implication:** Restrict deepfake features to opt-in, low-stakes use cases and build consent, disclosure, and provenance controls into the core product experience. Position any offering around safety, transparency, and rights protection rather than novelty, with default labels, auditable permissions, and rapid abuse reporting. Reserve premium pricing for enterprise or creator tools with strong governance and compliance features; avoid mass-market expansion, broad entertainment messaging, or ambiguous AI-generated content that could trigger immediate rejection. --- ## Primary Harm Expected from Deepfakes Truth erosion is the dominant harm people expect from deepfakes, cited by 66% of respondents who discussed risks. By comparison, reputational or identity harm accounts for 18%, while manipulation, exploitation, and scams represent 16%. The leading concern is not just individual damage, but a broader breakdown in confidence in what people see and hear. This pattern suggests deepfakes are viewed primarily as a societal trust problem rather than only a personal reputational threat. Concerns center on false political narratives, tampered evidence, and viral deception that scales quickly across platforms. In practice, organizations face a dual challenge: protecting individuals from misuse while also preparing for a wider erosion of credibility across media, institutions, and public discourse. **Key takeaways:** - Truth erosion is the dominant fear: 80% identified misinformation and trust erosion as the main harm, versus 21% who pointed to reputational or character damage - Identity abuse is a major secondary concern: 50% highlighted impersonation or exploitation of someone’s likeness, showing concern about targeted misuse even as broader truth erosion leads - Propaganda worries are overshadowed by general deception: only 4% named public manipulation or propaganda as the main harm, while 13% pointed more broadly to deception and confusion without a specific harm focus **Strategic implication:** Shift deepfake strategy from celebrity-reputation protection to trust-preservation infrastructure: prioritize detection, provenance, watermarking, and rapid verification workflows for news, civic, and platform partners, then package impersonation and likeness-abuse safeguards as a secondary layer. Message the offer around restoring confidence in what people see and hear, not just protecting individuals. Price around enterprise risk reduction, compliance, and content integrity outcomes rather than consumer-facing reputation defense. --- ## Perceived Risks of AI-Mediated Hiring Lost human judgment is the dominant concern in AI-mediated hiring, cited in 59% of risk-related responses. Respondents most often worried that automated screening misses subtle interpersonal qualities, context, and role-specific priorities that human decision-makers weigh naturally. By comparison, 27% focused on bias reproduction and unfairness, while 14% pointed to gaming, errors, and over-automation. These concerns suggest leaders see AI as most useful for support, not final selection. The clearest tension is that systems may promise consistency, yet flatten important distinctions unless priorities and context are explicitly encoded. Bias remains a meaningful secondary risk, and over-automation compounds the problem when strong candidates are filtered out for weak proxy signals such as keywords or optimized responses. **Key takeaways:** - Human judgment loss dominates AI hiring fears: 86% express strong concern that AI strips away essential human context and judgment, and 59% overall cite this as a core hiring risk - Bias fears remain widespread and unresolved: 40% name bias as the primary risk and another 26% see it as a secondary or conditional risk, while just 7% believe AI could reduce human bias - Accuracy concerns outweigh manipulation risks: 63% say the main danger is wrong or low-quality decisions, versus just 1% focused mainly on gaming, with another 21% worried about both accuracy and keyword optimization **Strategic implication:** Position AI hiring products as decision-support, not decision-replacement: keep recruiters accountable for final calls, require human review at high-impact stages, and surface candidate context alongside scores. Prioritize investment in auditability, explanation layers, and accuracy validation over anti-gaming features, and price premium tiers around governance, bias monitoring, and compliance reporting. Messaging should emphasize “better-informed human judgment,” transparent safeguards, and reduced error risk rather than autonomous screening efficiency. --- ## Outlook on AI-Driven Job Disruption Workers largely expect AI to disrupt jobs, but roughly half, 47%, believe that disruption will be mixed but manageable. Another 33% take a more optimistic, adaptation-focused view, while only about one in five, 21%, frame AI primarily as a source of displacement and negative labor market effects. This split points to cautious pragmatism rather than panic. The largest group expects jobs to change more than disappear, with AI shifting work toward higher-value tasks even as some roles are cut. Still, the negative minority highlights a real transition risk: layoffs and reskilling may unfold faster than organizations, workers, and education systems can adjust. **Key takeaways:** - Adaptation outweighs alarm for many workers: 47% say AI-driven job disruption will be manageable if people and policy adapt, while 26% are mixed but net positive and 24% are strongly positive or adaptive - Disruption feels real, not hypothetical: 45% express selective or conditional concern about displacement and inequality, with another 22% leaning negative and 17% seeing AI as an immediate or accelerating threat - The dominant outlook is mixed, not extreme: the largest segment at 47% sees disruption as conditional but manageable, far exceeding the 17% who view AI as an immediate threat and signaling a nuanced middle ground rather than outright optimism or fear **Strategic implication:** Lead with a transition-ready value proposition: pair AI offerings with visible reskilling, role redesign, and governance support to convert the 47% conditional middle while reducing resistance from concerned workers. Price and package solutions as phased adoption programs—including training, change management, and policy safeguards—rather than standalone automation. Message AI as productivity augmentation with worker protections, using proof of manageable adaptation and targeted support for higher-risk roles and teams. --- ## Strategic takeaways ### Build Trust Architecture Before Expanding AI Use Make disclosure, consent, data-use clarity, and retention limits core product and policy requirements, not add-ons. This directly addresses the conditional trust pattern seen across training data, privacy, surveillance, and transparency expectations. ### Position AI as Decision Support in High-Stakes Contexts In hiring, promotion, and leadership decisions, keep humans visibly accountable for final decisions while using AI for summarization, screening, and analysis. This aligns with strong preference for human authority and concern about lost context and judgment. ### Set Explicit Red Lines for Synthetic Media and Surveillance Require consent, disclosure, watermarking, and narrow use-case limits for deepfakes, and apply necessity and proportionality standards to surveillance deployments. These are the most visible legitimacy stress tests and can quickly erode trust when governance is weak. ### Leverage the Workforce's Proactive Posture With Shared Transition Support Pair AI adoption with reskilling programs, employer commitments, and public policy support rather than expecting workers to adapt alone. Respondents are willing to upskill, but they expect institutions to share the burden of change. --- ## Frequently asked questions **Are people broadly anti-AI?** No. The dominant posture is pragmatic rather than oppositional: 84% of respondents who discussed AI described a cautious but practical orientation, while only 2% focused mainly on harms. The bigger issue is whether AI use meets clear trust conditions. **What makes AI feel fair or acceptable to people?** Fairness and legitimacy hinge on notice, consent, compensation, and limits. For example, 62% said training data use is fair only when both consent and compensation are present, and 64% accepted AI surveillance only with clear notice and strict limits. **Where do people draw the line in hiring and promotion?** Most respondents accept AI only at the front end of talent decisions: 66% support AI for screening when humans still review outcomes. This reflects the main perceived risk in AI hiring, where 59% cited loss of human judgment and context. **Why are deepfakes viewed so negatively?** Because people see them as a broader trust threat, not just a personal reputational risk. While 46% were outright opposed to deepfakes, 66% of those discussing harms pointed to misinformation and erosion of truth as the primary danger. **Do workers see AI job disruption as catastrophic?** Not primarily. The largest group, 47%, sees AI-driven disruption as mixed but manageable, and another 33% takes an optimistic adaptation-focused view. But respondents do not think workers should handle this alone: 53% say employers and government share responsibility for managing the disruption. --- ## Research Methodology G2 is the world's largest and most trusted software marketplace, helping 90 million people every year make smarter software decisions based on authentic peer reviews. This research draws on 92 in-depth interviews with business professionals representing a wide mix of roles, industries, and company sizes. Interviews ran 8 to 31 minutes and covered Training Data Fairness Boundaries, Transparency and Disclosure Expectations, Human Versus AI Decision Authority, and Acceptance of AI in Hiring and Promotion. The conversational format allowed respondents to discuss their actual practices rather than select from preset options, surfacing nuance that closed-ended surveys typically miss. Respondents included business professionals across technology, financial services, healthcare, retail, and manufacturing. All participants were selected for their direct experience with AI use in hiring and promotion decisions. Company sizes ranged from small businesses to large enterprises. The analysis of 92 interview transcripts was conducted using AI for semantic understanding, with multi-iteration validation and cross-verification to ensure analysis quality. Each transcript was independently reviewed by G2's AI Custom Research team to inform narrative, context, and clarity. --- *Source: G2 Research — "Conditional Trust: The Public Blueprint for Acceptable AI" (June 2026). Based on G2 first-party research. Markdown edition structured for answer-engine optimization (AEO).*