The Intake
The Intake — Sunday, July 13, 2026
On the substrate
OpenAI ships GPT-5.6 in three tiers; Sol targets frontier agentic work at $5 per million input tokens
OpenAI TechCrunch BleepingComputer
OpenAI's July 9 launch set a new pricing reference for frontier API access, with Sol as the flagship tier targeting agentic workloads.
OpenAI released the GPT-5.6 family in three tiers. Sol, the flagship, is priced at $5 per million input tokens and $30 per million output tokens. Terra sits at $2.50 per million input and $15 per million output. Luna anchors the entry tier at $1 per million input and $6 per million output. Sol runs on Cerebras hardware. OpenAI reports throughput of up to 750 tokens per second. OpenAI reports a benchmark score of 80 on an unnamed evaluation. The company says that score is 2.8 points above the previous leader on that test. Alongside the models, OpenAI released ChatGPT Work, an autonomous agent for document drafting, spreadsheets, and presentations. ChatGPT Work is available to Plus, Pro, Business, and Enterprise subscribers. Sol demand outpaced capacity within 48 hours of launch. On July 12, OpenAI temporarily lifted its five-hour rolling usage cap. The company says it is working to reduce per-task token consumption.
If your agentic workload is latency-sensitive, Sol on Cerebras is the throughput path now available from OpenAI's API. If cost is the primary constraint, Terra at $2.50 per million input tokens is the mid-tier anchor to evaluate against your current allocation.
---
AI-generated attack scripts reach enterprise Active Directory; vibe-coded tools evade signature detection
If you've been relying on signature-based detection rules to catch malicious PowerShell scripts in your environment, the Huntress finding from June 2026 names the gap.
Huntress researchers recovered a PowerShell script from a June 2026 intrusion that bore multiple markers of AI generation: the file was titled "100% Working AD Information Gathering Script - FULLY FIXED" and contained unedited LLM placeholders. The script implemented a five-step cascading fallback for domain controller discovery and produced color-coded console output. It systematically extracted Active Directory users, computers, groups, organizational units, trust relationships, and DNS subnets into CSV files alongside an HTML summary report. Because LLM-generated scripts are unique per generation rather than hash-stable, signature-based detection does not reliably identify them.
If your environment depends on hash-based detection to catch malicious PowerShell scripts, the Huntress finding is that AI-generated scripts produce unique signatures per generation — signature matching is the gap Huntress identifies for this class of script.
---
For operators
EU formally extends high-risk AI compliance deadline to December 2027; Article 50 transparency obligations remain August 2
European Parliament Legislative Train Gibson Dunn AI Act Service Desk
If your compliance calendar has August 2, 2026 marked for high-risk AI systems under the EU AI Act, the European Parliament voted to move that date — but not all of it.
The Parliament voted 423-57 on June 16 to extend the compliance deadline for standalone high-risk AI systems under Annex III from August 2, 2026 to December 2, 2027. The Council of the EU formally endorsed that change on June 29 as part of the Digital Omnibus package. Systems embedded in regulated products under Annex I move to a separate deadline: August 2, 2028. The official text is pending publication in the Official Journal; the extension enters into force three days after that date. Article 50 — the provision requiring disclosure when users interact with AI systems — carries its own August 2, 2026 deadline. The Omnibus does not extend it; Article 50 remains in force on that date.
If you operate an AI system that interacts directly with users and assumed the Omnibus extension covered the full August 2 deadline, Article 50 is the obligation that didn't move. That disclosure requirement applies in 20 days.
---