IOSG argues that private AI is no longer a niche technical preference but an emerging requirement for enterprises and power users that do not want proprietary data, internal workflows, or high-value judgment calls exposed to model providers. In its report, the firm lays out how the privacy problem starts the moment a prompt leaves a user’s device and reaches a server in plaintext, and why contractual protections such as zero-data-retention terms can only go so far. The piece links that risk to corporate restrictions on ChatGPT, shadow AI leaks, and a series of legal cases in which user chats became discoverable evidence. The report also maps the trade-offs across today’s privacy stack, from contract-based retention promises and OHTTP relays to trusted execution environments, end-to-end encryption, fully homomorphic encryption, and local inference. Its central case study comes from Bridgewater’s AIA Labs and Thinking Machines, which showed that a fine-tuned open model, Qwen3-235B, beat frontier models on both accuracy and cost in financial judgment tasks. IOSG’s conclusion is narrow but clear: for execution-heavy agent workflows, trust-based setups still dominate because tool calls expose plaintext to downstream services; for high-value strategic reasoning and domain-specific alpha, verified private infrastructure around open models is becoming a practical path.

A new report from IOSG argues that the core debate in AI is shifting from model capability alone to a harder question: who gets to see the data, and whether privacy claims can actually be verified. The piece points to a string of examples showing why that matters. Palantir CEO Alex Karp said companies are paying a token premium to frontier labs while letting proprietary knowledge leak out through plaintext requests. Wall Street banks restricted ChatGPT use within months of its launch, Samsung banned generative AI across its network after engineers exposed chip source code, and court orders later forced OpenAI to retain and disclose consumer chat records in litigation. The report maps the current privacy stack, from contractual zero-data-retention and anonymous relays to trusted execution environments, end-to-end encryption, fully homomorphic encryption and local inference. It argues that verifiable privacy is still mostly limited to open models, because frontier labs have little incentive to expose model weights or serving code. At the same time, the economics are changing: enclave-based inference is getting cheaper, and in some cases can match or undercut plaintext API pricing. IOSG also highlights a June 30 case from Bridgewater-backed AIA Labs and Thinking Machines, where a fine-tuned open model beat frontier systems on both accuracy and cost in financial tasks. The report’s broader point is that private AI remains incomplete, especially for agentic workflows and tool use, but it is no longer hypothetical.

Paradigm’s annual conference and hackathon, Frontiers, is scheduled for Oct. 12-14, 2026, in San Francisco, according to a post on X by Georgios Konstantopoulos, the firm’s general partner and chief technology officer. The event will run on an invite-only basis and will not include booths or sponsors. Topics listed for the gathering include stablecoins, artificial intelligence, robotics, and post-quantum cryptography. Paradigm said the event is aimed at developers and technologists working at the edge of these fields. Applications are now open.

U.S. President Donald Trump said he still believes the United States could reach an agreement with Iran, while also saying Tehran has resumed contact with Washington. Speaking to reporters at the White House on July 13 local time, Trump said Iran wants a deal and has re-engaged with the U.S. He also said the U.S. military would continue heavy strikes on Iran to sharply reduce its ability to affect shipping through the Strait of Hormuz. In the same remarks, Trump said the U.S. is restoring a blockade "targeted only at Iran," adding that ships doing business with Iran would not be allowed to pass, while other countries and vessels could continue normal transit. The remarks were cited by CCTV International News and carried by BlockBeats on July 14.

OpenAI has published a new prompting guide for GPT-5.6 Sol, and its central recommendation runs against a year of prompt-writing habits: stop stuffing system prompts with extra instructions. The company says the better approach is outcome-first prompting—state the desired result, define success criteria, specify stopping conditions, and leave out process detail the model already handles on its own. According to OpenAI, internal coding-agent tests showed that leaner system prompts improved evaluation scores by about 10% to 15%, while reducing total token use by 41% to 66% and lowering costs by 33% to 67%. The guide argues that repeated rules, style notes that do not change behavior, and examples that add no functional value now act as noise rather than support. The update also marks a clear shift from the GPT-5 guide released in August 2025, which favored heavier scaffolding such as XML persistence blocks, detailed context-gathering templates, and narrated tool preambles. GPT-5.6, OpenAI says, needs fewer rails. The guide also introduces the text.verbosity parameter and a new section on Programmatic Tool Calling for workflows where code can filter, batch, or aggregate outputs before handing compact results back to the model.

Dragonfly partner Haseeb said on X that DeFi has not entered the “hacker doomsday” some had feared, even as models including GLM 5.2, Fable and GPT 5.6 are already live and being used by attackers. Citing chart data, he said 2026 DeFi losses are running at an annualized $1.89 billion based on year-to-date figures and the current pace. Total losses so far this year stand at about $986 million, below 2025 levels and still within the historical range. Haseeb said the more meaningful shift is in attack composition: the number of hacks has increased, but the size of each attack has fallen faster. He added that attackers are focusing more on smaller protocols and abandoned projects, while larger protocols have completed more security hardening. In his view, overall fund safety across DeFi has not deteriorated in a clear way.

Cap, a Franklin Templeton-backed stablecoin protocol, has reduced its promised Stabledrop user reward to $4.2 million from roughly $12 million, prompting a public apology from founder Benjamin Sarquis Peillard. The project said the payout will be funded entirely by proceeds from its June token sale, which raised $4.2 million after Cap sold 5% of token supply at a $75 million floor, far below the earlier assumption of a $250 million valuation and a 10% sale. The revised distribution now applies only to certain Frontier users who bought Pendle yield tokens without holding the matching principal tokens, while cUSD holders and Pendle liquidity providers are excluded. Sarquis Peillard also rejected accusations that he routed rewards to a wallet tied to his former employer, saying the address belongs to a former colleague and received no QiDao treasury funding beyond gas fees. CAP has fallen about 20% over the past week, while Cap’s total value locked dropped to about $230 million from a July 8 peak of $262 million, according to DefiLlama.

Sablier Labs, the company behind token-streaming and vesting infrastructure, has halted active product development and moved into maintenance mode through June 2028, according to co-founder and CEO Paul Berg. The company said existing streams, vesting plans, and airdrops will continue to operate because Sablier’s smart contracts are deployed onchain and do not depend on the company remaining in business. Starting July 13, 2026, the official interface stopped accepting new vesting streams and airdrops with end dates beyond June 2028, and it no longer allows open-ended payment streams. Berg linked the decision to a sharp drop in usage and revenue in the first quarter of 2026, even though the company shipped more features than in any prior quarter. He cited delayed token launches as crypto markets weakened and lower barriers for rivals as AI-assisted coding made replication cheaper. Sablier also moved up the license conversion for its main EVM smart contracts, switching immediately from Business Source License 1.1 to GPL on July 13, 2026, instead of waiting until July 1, 2029. The company said more than 345,000 Ethereum addresses have used the protocol across over 837,000 transactions and more than 547,000 vesting plans, airdrop claims, and payment streams on 30-plus EVM chains and Solana.