Immune checkpoint therapies have transformed immunotherapy but remain dominated by biologic agents characterized by prolonged receptor occupancy and limited pharmacologic controllability. Although multiple classes of immunomodulatory therapeutics exist, synthetic modalities capable of directly targeting extracellular immune checkpoint protein-protein interaction interfaces while enabling controllable immune modulation remain comparatively underexplored. Here, we report an AI-guided strategy for discovering cyclic peptide antagonists of the costimulatory receptor CD28. The lead peptide, CIP-3, binds the CD28 extracellular domain with nanomolar affinity and disrupts CD28-ligand interactions. In primary human immune systems, CIP-3 suppresses CD28-dependent T-cell activation without intrinsic agonist activity and exhibits rapid pharmacologic reversibility, enabling exposure-dependent control of immune signaling. In a T-cell transfer model of chronic colitis, CIP-3 confers dose-dependent therapeutic efficacy and reduces systemic inflammatory cytokines. CIP-3 also suppresses cytokine production across independent healthy donors and patient-derived PBMCs from individuals with ulcerative colitis with efficacy comparable to a benchmark anti-CD28 biologic. Together, these findings demonstrate the potential of AI-designed cyclic peptides as a controllable synthetic modality for immune checkpoint modulation.