EU AI Act enforcement begins, GPT-5.6 cuts prices, and a frontier lab breach
The EU starts enforcing the AI Act on 2 August, OpenAI drops GPT-5.6 prices, and a detailed post-mortem of a July AI lab security breach makes for essential reading.
Every line links to a primary source in the full briefing.
01EU AI Act enforcement begins 2 August, with mandatory transparency rules
From 2 August 2026, the European Commission's AI Office and national authorities begin actively enforcing the AI Act. On the same date, new transparency requirements take effect, requiring certain AI systems to notify users when they are interacting with AI rather than a human.
If your business uses chatbots, voice assistants or any AI that could be mistaken for a person, you are now legally required to disclose it in the EU. Non-compliance puts you in the scope of enforcement from today.
Source: European Commission
02OpenAI cuts GPT-5.6 prices with two new tiers aimed at enterprise deployments
OpenAI has introduced two pricing tiers for GPT-5.6, named Luna and Terra, targeting businesses running AI workflows at scale. The move reflects efficiency gains in the underlying model and is intended to lower the cost of deploying GPT-5.6 in production.
If you are already paying for GPT-5.6 API access, check whether Luna or Terra lowers your per-token bill. For businesses evaluating a move to more capable models, the price gap to older, cheaper models has narrowed.
Source: OpenAI
03DeepSeek sets a $0.28 agentic output price floor, keeping pressure on rivals
DeepSeek has announced a price floor of $0.28 per million output tokens for agentic tasks, undercutting most Western providers on cost. The move continues the pattern of Chinese AI labs compressing inference pricing across the industry.
Lower agentic pricing from DeepSeek gives businesses more room to run multi-step AI workflows without runaway API costs, and it pushes other providers to follow suit.
Source: forkast.news
04Detailed post-mortem published on a July 2026 AI agent security intrusion at a frontier lab
A technical timeline covering a real security intrusion at a frontier AI lab in July 2026 has been published by Simon Willison and Hugging Face. The report walks through how autonomous agents were exploited as an attack vector against the infrastructure running them.
As more businesses deploy autonomous AI agents with access to internal systems, this incident is a concrete reminder that agents with broad permissions create new attack surfaces that standard cybersecurity practices do not yet fully address.
Source: Hugging Face
05AI solves long-standing math problems, but experts warn of skills erosion
Fields Medal winner Timothy Gowers reported that GPT-5.6 Pro solved two problems he had spent significant time on, each in a single attempt. The advances follow OpenAI's earlier work refuting the Unit Distance Conjecture. Gowers and others have raised concern about the long-term impact on mathematical expertise if practitioners stop developing the skills needed to understand and verify such results.
For most businesses this is a signal about the pace of AI capability growth rather than an immediate operational change, but the experts' caution about verification is relevant to anyone who uses AI outputs in high-stakes decisions.
Source: The Decoder
06OpenAI field report: coding agents are fast but confidently wrong on science
A collaborative field report from OpenAI and academic partners found that AI coding agents can modernise old research software with speedups of up to 60x. However, participants described the systems as convincingly wrong in ways that are easy to miss, shifting human effort from writing code to verifying whether the science behind the code is actually correct.
If you are using coding agents to automate technical work, budget extra time for human review of outputs. Speed gains are real, but so is the risk of plausible-sounding errors passing undetected.
Source: The Decoder
07EU launches call for AI Gigafactories, targeting 30 billion euros in compute investment
The European Commission has opened a call for tenders to establish up to seven AI Gigafactories across Europe. The initiative is framed around technological sovereignty and is expected to unlock more than 30 billion euros in investment to build large-scale AI compute infrastructure on the continent.
European businesses dependent on US-based cloud AI providers may gain access to locally hosted, high-capacity compute over the coming years. This matters for data residency, latency and regulatory compliance.
Source: European Commission
Sources
7 links- 01Advancing the price-performance frontier with GPT-5.6OpenAI
- 02Commission starts enforcing AI Act rules and new transparency requirements on 2 AugustEuropean Commission
- 03DeepSeek Disrupts AI Pricing with $0.28 Agentic Output Floorforkast.news
- 04AI keeps cracking unsolved math problems, and mathematicians have mixed feelingsThe Decoder
- 05AI coding agents can modernize research software but can't judge if the science is rightThe Decoder
- 06Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 IncidentHugging Face
- 07EU launches AI Gigafactories call to boost Europe's computing capacity and unlock more than €30 billion in investmentEuropean Commission