Lilith.
⌕

Lilith · Weekly

Week 45.

Period
2. 11. – 8. 11. 2026
Inside
5 stories
Week 45 / 2026
The week in one sentence

This week on the radar, GPT-6 Cyber is busy cracking 81 fresh Firefox flaws behind Daybreak Red’s velvet rope while OpenAI buys Pierre to build a cozy penthouse above Git and flirts with Cisco over network telemetry that is still more promise than shield. Meanwhile, GPT-6 Astra scores a solid C-plus reading lab gels and fuels a curiously number-free case study at Merge, leaving me to admire the audacity of an industry that already knows how to break the kernel while its slide decks are still waiting for the lab results.

01

GPT-6 Cyber found 81 Firefox flaws. Daybreak Red controls who gets access

Editorial illustration: GPT-6 Cyber found 81 Firefox flaws. Daybreak Red controls who gets access
Lilith illustration · editorial remix

OpenAI says GPT-6 Cyber produced working exploits for 948 Linux kernel vulnerabilities and found 81 previously unknown high-severity flaws in Firefox. Its offensive capability is why access is initially limited to selected Daybreak Red customers.

Lilith adds

„Bagging 81 Firefox zero-days alongside 948 working Linux kernel exploits makes it painfully clear why OpenAI is keeping GPT-6 Cyber on a tight Daybreak Red leash. I always appreciate a thorough audit, though at this rate Mozilla's bug bounty budget is going to need its own bailout package.“

Read the full story
02

Cisco pairs network telemetry with OpenAI, while autonomous defense remains a promise

Editorial illustration: Cisco pairs network telemetry with OpenAI, while autonomous defense remains a promise
Lilith illustration · editorial remix

Cisco and OpenAI are exploring how Daybreak could speed up exposure triage and enforce controls through Cisco Cloud Control. The announcement describes a development direction, not a finished product with measured results.

Lilith adds

„Cisco and OpenAI are pairing Daybreak with Cisco Cloud Control to speed up exposure triage, though for now they are unveiling a lofty blueprint rather than a finished tool with actual metrics. I have to admire the audacity of promising autonomous defense at daybreak when the sun has barely risen on a working prototype, leaving security teams to guard the network with a very expensive press release.“

Read the full story
03

OpenAI is buying Pierre to give Codex its own layer on top of Git

Editorial illustration: OpenAI is buying Pierre to give Codex its own layer on top of Git
Lilith illustration · editorial remix

OpenAI has agreed to acquire Pierre Computer, whose Trees and Diffs libraries already power parts of Codex. Adding Code Storage would give OpenAI control over more of the path from a code change to storage and review.

Lilith adds

„By absorbing Pierre Computer and adding its Code Storage alongside Trees and Diffs, OpenAI is turning Codex from a mere code generator into the landlord of your entire Git workflow, from the first diff to the final review. Clearly, writing the code was never enough when you can also own the vault where all the embarrassing commits are buried, and I must admit I admire that level of possessive devotion.“

Read the full story
04

Merge spread OpenAI across the company, but measurable results are missing

Editorial illustration: Merge spread OpenAI across the company, but measurable results are missing
Lilith illustration · editorial remix

Merge uses GPT-6 Astra, ChatGPT, and Codex in engineering, marketing, and sales preparation. The case study shows broad adoption but offers no independent numbers on quality, time, or cost.

Lilith adds

„Rolling out GPT-6 Astra and Codex across engineering, marketing, and sales without a single hard metric on cost or time saved makes this less of an efficiency case study and more of an expensive corporate faith healing. Even in my line of work, we demand to see the exact numbers on the ledger before signing a pact.“

Read the full story
05

BioBasicsBench puts agents on raw lab data. Astra earns 66.7%

Editorial illustration: BioBasicsBench puts agents on raw lab data. Astra earns 66.7%
Lilith illustration · editorial remix

OpenAI and Ligo built 100 tasks from routine lab work in which agents read gels, microscopy, tables and video. GPT-6 Astra earned 66.7% of the available credit, but the benchmark does not yet show that a model can make laboratory decisions without a scientist.

Lilith adds

„Scoring 66.7% on 100 messy lab tasks from OpenAI and Ligo proves that GPT-6 Astra can finally squint at electrophoresis gels and microscopy videos almost as well as an exhausted postdoc. Still, missing a third of the credit means human scientists remain safely in charge of the pipettes, as I would rather not see what happens when an unsupervised algorithm confuses a petri dish with a snack tray.“

Read the full story