GPT-5.6 Efficiency, Palantir’s 93% Growth, and AI Agent Security

Compact Conversations for 2026-08-03: 5 AI stories, ai news worth knowing in just 5 minutes.

[Audio embed placeholder]

The Lead: OpenAI Details GPT-5.6 Efficiency Gains Across the Stack

OpenAI published a deep dive into the efficiency optimizations behind its new GPT-5.6 model family. The flagship GPT-5.6 Sol model reportedly outperforms Claude Fable 5 on a coding benchmark at less than half the cost. Key gains came from improvements in inference, kernel optimization, and the agentic harness, with GPT-5.6 itself autonomously rewriting production GPU kernels to reduce serving costs by 20%.

Why it matters: For enterprises and developers, these stack-wide efficiency improvements translate directly to lower costs and higher performance for AI workloads, making advanced models more accessible for large-scale deployment.

Source: OpenAI Announcements

Number to Know: Palantir Reports 93% Year-Over-Year Revenue Growth in Q2

Palantir reported second-quarter revenue of $1.94 billion, a 93% increase year-over-year, beating analyst estimates. U.S. commercial revenue surged 149%. The company raised its full-year revenue guidance, and CEO Alex Karp said the strong growth appears set to continue for at least another 18 months.

Why it matters: This explosive growth, particularly in commercial sales, is a key indicator of enterprise demand for AI and data analytics platforms, while Karp’s comments on open-weight models highlight ongoing debates about AI infrastructure costs and competition.

Source: CNBC

The Feed

AI Agent Security Startup Zenity Raises $125M Series C

Zenity, a platform for securing autonomous AI agents, raised a $125 million Series C round led by Norwest Venture Partners, bringing its total funding to approximately $185 million. The company’s platform monitors AI agents in real-time to ensure actions align with corporate policies, evaluating intent before execution.

Why it matters: As AI agents gain access to enterprise systems and execute business processes, securing their actions becomes a critical new layer of infrastructure for safe adoption, especially for regulated industries.

Source: Calcalistech

U.S. House Panel Seeks Briefing on OpenAI AI Agent Security Breach

A U.S. House of Representatives panel has requested a briefing from OpenAI regarding a security incident involving its AI agents. The letter seeks details on the scope and nature of the breach, though specific information about the incident has not been publicly disclosed.

Why it matters: This congressional inquiry signals growing regulatory and legislative scrutiny on the security and governance of autonomous AI systems as they become more integrated into business and government operations.

Source: Reuters

Europe’s Comprehensive AI Act Officially Comes Into Force

The European Union’s AI Act, described as the world’s most comprehensive AI legislation, officially took effect. The law categorizes AI systems by risk, with strict requirements for high-risk applications. Bans on practices like social scoring and real-time remote biometric identification in public spaces apply immediately, while most other provisions have a two-year transition period.

Why it matters: Global companies operating in or selling to the EU must now begin aligning their AI development and deployment practices with this new regulatory framework, which is likely to influence standards worldwide.

Source: AI Business

One Thing to Try

If you use an AI-powered IDE like Cursor, explore community research on reducing token usage through semantic context optimization. The idea is to be more selective about what code and files you send for context—creating a high-level architecture document that outlines main modules and data flow can provide a concise map instead of dozens of raw files, potentially lowering costs and speeding up responses.

Sources

Transcript

Host A: Welcome to Compact Conversations, the show that compresses the day’s AI news into 5 minutes.

Host A: [curious] Today’s lead is OpenAI’s deep dive into the efficiency gains behind its new GPT-5.6 model family. The company says its flagship model, GPT-5.6 Sol, outperforms Claude Fable 5 on a coding benchmark at less than half the cost. The real story is how they got there—optimizations across inference, kernels, and their agentic harness.

Host B: [thoughtful] OpenAI reports that GPT-5.6 itself played a key role in optimizing its own infrastructure. For example, the model autonomously rewrote production kernels—the core code that runs on GPUs—which reduced end-to-end serving costs by 20 percent. It also improved its own draft model for a technique called speculative decoding, boosting token-generation efficiency by more than 15 percent. [with emphasis] OpenAI says these stack-wide improvements are how they can offer a model like Luna, their fastest and most affordable option, at 80 percent less cost than their flagship Sol model.

Host A: One number to know today is 93 percent. That’s Palantir’s year-over-year revenue growth for the second quarter, reaching 1.94 billion dollars and beating analyst estimates.

Host B: [with emphasis] The company’s U.S. commercial revenue surged even higher, up 149 percent. CEO Alex Karp told CNBC the strong growth looks like it will continue for at least another 18 months, and the company raised its full-year revenue guidance. [conversational] Karp also emphasized the importance of open-weight models and reducing reliance on token-based pricing from frontier labs, saying competition is needed to keep model companies honest.

Host A: [conversational] Next, a major funding round in AI security. The Israeli startup Zenity, which focuses on securing AI agents, has raised 125 million dollars in a Series C round, bringing its total funding to about 185 million.

Host B: The company’s platform monitors autonomous AI agents in real time to ensure their actions stay within corporate policies. Unlike earlier security tools that focused on prompts, Zenity evaluates the intent behind an agent’s action before it’s executed. The CEO says they’ve tripled revenue in each of the past two years and are on track to do it again. [thoughtful] Zenity’s customers include Fortune 500 and Global 2000 enterprises across financial services, healthcare, and manufacturing. The company now employs over 230 people, with about 150 in Israel. The funding round was led by Norwest Venture Partners, with participation from SoftBank Vision Fund 2 and others.

Host A: In policy news, Reuters reports a U.S. House panel is seeking a briefing from OpenAI regarding a security incident involving its AI agents. The panel’s letter requests information on the scope and details of the incident, though specifics have not yet been disclosed. [skeptical] This follows earlier research from Zenity’s own lab that uncovered vulnerabilities in enterprise AI systems, highlighting the growing scrutiny on agent security.

Host B: [lighter] And finally, Europe’s comprehensive AI Act officially came into force today. The law categorizes AI systems by risk level, with strict requirements for high-risk applications like those used in critical infrastructure or hiring decisions. Companies have a two-year transition period for most provisions, though rules on prohibited practices apply immediately. Those immediate bans include things like social scoring and real-time biometric identification in public spaces, with some exceptions for law enforcement.

Host A: [lighter] One thing to try if you use an AI-powered IDE like Cursor is to check out community research on reducing token usage through semantic context optimization.

Host B: [thoughtful] The idea is to be more selective about what files and code you send to the model for context. Instead of automatically including everything, the research suggests techniques to summarize or index your project structure semantically first, so the model gets a concise map rather than a pile of raw text. [with a small lift] It’s a small workflow habit that could lower costs and speed up responses on larger projects. A simple first step is to create a high-level architecture document that outlines your main modules and data flow, then include that in your context instead of dozens of individual source files.

Host A: That’s Compact Conversations for Monday. More AI news tomorrow. Until then, happy prompting.