OpenAI rolls out GPT‑5.3 System Card, revises DoD pact to curb US surveillance, and

OpenAI unveiled the GPT‑5.3 Instant System Card on its website, detailing a new “instant” mode that reduces latency for real‑time applications and adds a concise safety checklist for developers (OpenAI). The card also announces Codex‑Spark, the first GPT‑5.3 model running on Cerebras wafer‑scale chips instead of Nvidia GPUs, promising higher throughput and lower latency for interactive coding tasks (InfoQ). OpenAI is rolling the Codex‑Spark preview out to ChatGPT Pro users, allowing early experimentation while the company expands Cerebras‑based datacenter capacity (InfoQ).

In parallel, CEO Sam Altman announced an amendment to OpenAI’s Department of Defense contract that explicitly bars the use of its AI systems for mass surveillance of U.S. persons. The revised language cites the Fourth Amendment, the National Security Act of 1947, and the FISA Act of 1978, stating the AI “shall not be intentionally used for domestic surveillance of U.S. persons and nationals” (Engadget). Altman shared the change in an internal memo posted on X, emphasizing that the Department understands the limitation to prohibit deliberate tracking, monitoring, or surveillance of Americans (Engadget).

The amendment follows growing pressure on AI firms to police government misuse of generative models. By embedding constitutional references directly into the contract, OpenAI aims to create a legally enforceable barrier against domestic spying, a step it says will “make that point especially clear” to the Defense Department (Engadget). The company did not disclose whether the clause will trigger penalties or audits, but the language suggests a formal compliance obligation.

OpenAI’s hardware shift to Cerebras chips aligns with the system‑card’s claim of “ultra‑fast coding speeds,” a performance boost that could make the model attractive for defense‑grade software development where latency is critical (InfoQ). The move also signals a diversification away from Nvidia, a trend noted in recent coverage of OpenAI’s broader product updates (ZDNet, Wired). By coupling faster hardware with stricter usage policies, OpenAI appears to be positioning GPT‑5.3 as both a high‑performance tool and a responsibly governed asset for government partners.

Analysts have not yet quantified the commercial impact of the DoD amendment, but the combined rollout of the system card, Codex‑Spark, and the surveillance restriction marks the most significant policy and product shift for OpenAI since the GPT‑5 series launch (VentureBeat). The company’s next steps will likely focus on scaling Cerebras capacity and monitoring compliance with the new contractual language as it expands its foothold in both enterprise and defense markets.