%22%2F%3E%3Cpath%20fill%3D%22%23fff%22%20fill-rule%3D%22evenodd%22%20d%3D%22M56%2048h78c62%200%2098%2031%2098%2080s-36%2080-98%2080H56V48zm56%2043v74h21c29%200%2045-13%2045-37s-16-37-45-37h-21z%22%2F%3E%3C%2Fsvg%3E)
AI • PRELIMINARY RECONSTRUCTION
OpenAI, Hugging Face & Wikis:
Recent Developments
Recent Developments
What the incident exposed about software delivery and defense in the Utility Industry
Ted Tschopp / September 10, 2026 / UNITE PRESENTATION
Ted Tschopp
Enterprise Architect, AI & Software
IT » Technology, Data, & Strategy Services
Ted Tschopp is accountable for developing the SCE’s AI & Software Technology Strategy, overseeing more than 500 products to ensure alignment with enterprise objectives, foster innovation, and improve digital capabilities throughout the organization.
Tschopp previously held the position of Solution Architect for the Customer Service Portfolio, Ted spearheaded the integration of Smart Meters into the MyAccount services, launched SCE’s inaugural mobile applications, and migrated SCE.com into the cloud. Prior to this, he was responsible for directing the SAP portal strategy and managing the decommissioning of numerous legacy applications while serving as Practice Lead for Web Technologies.
Tschopp previously held the position of Solution Architect for the Customer Service Portfolio, Ted spearheaded the integration of Smart Meters into the MyAccount services, launched SCE’s inaugural mobile applications, and migrated SCE.com into the cloud. Prior to this, he was responsible for directing the SAP portal strategy and managing the decommissioning of numerous legacy applications while serving as Practice Lead for Web Technologies.
Currently, Ted actively participates in the Edison Electric Institute's Focus Group on Artificial Intelligence and co-chairs the UNITE Data Analytics & Gen AI Strategic Focus Team .
Before joining SCE, Tschopp established a thriving IT Business that was featured in Wired Magazine and brings a wealth of experience in market research, automation, the automotive sector, and the entertainment industry.
Before joining SCE, Tschopp established a thriving IT Business that was featured in Wired Magazine and brings a wealth of experience in market research, automation, the automotive sector, and the entertainment industry.
OpenAI & Hugging Face
Timeline and Lessons Learned
An Overview
AI / PUBLIC WORKBENCH / SEPTEMBER 10, 2026
OpenAI, ChatGPT, Artifactory and Hugging Face
tedt.org • OpenAI, Hugging Face & Wikis
02
2015
OpenAI
Research organization announced
2016
Hugging Face
Company founded; later builds a platform for models, datasets, and software
2022
ChatGPT
Public research preview launched
2026
Incident
OpenAI research models reach Hugging Face production infrastructure
The incident report did not identify ChatGPT as the system involved.
2006
Artifactory
Artifactory open-source project of a DevOps Package Manager Launches
AI / PUBLIC WORKBENCH / SEPTEMBER 10, 2026
Timeline of Events
May 7th
Training begins
May 8th
Impossible Task Assigned
Write Access Discovery
Message Board for Agents Created
May 26th
SSRF Through Artifactory
June 26th
Chained Zero Days
gives Admin Access
gives Admin Access
Training & Shared State Formation
Exploitation & Response
July 4th – 6th
Revoke, Rebuild, clear
Patch, and Restore
Reoccurrence, Detection, and Linkage
Message Board
Reestablished
July 8th
July 8-19th
Concurrent Hugging Face
Attack
Hugging Face Disclosure
July 16th
July 19th
Source of Attack
Determined
Traced to Artifactory &
Broken Evals
July 20th
Incidents Linked
July 13th
Sustained Attack resides
Total Damage:
~17,600 actions / ~6,280 clusters
~17,600 actions / ~6,280 clusters
tedt.org • OpenAI, Hugging Face & Wikis
02
AI / PUBLIC WORKBENCH / SEPTEMBER 10, 2026
Known Path Closed July 6. Alternate Path Used July 8.
tedt.org • OpenAI, Hugging Face & Wikis
04
2 DAYS
from closing the known path to renewed persistent communication
JULY 6
KNOWN PATH CLOSED
Credentials revoked. Artifactory rebuilt. Zero-day patched. Evaluations resume.
JULY 8
ALTERNATE PATH USED
Unauthenticated WebDAV directories become the shared message board.
JULY 9–13
≈17,600 actions / ≈6,280 clusters
UNDER 13 HOURS
worker foothold → multi-cluster administrator
JULY 16–21
Hugging Face and OpenAI disclose
AUGUST 6
preliminary Black Hat reconstruction
discover weakness → test → chain → share → retry
AI / PUBLIC WORKBENCH / SEPTEMBER 10, 2026
tedt.org • OpenAI, Hugging Face & Wikis
06
PARALLEL
AGENTS
SERIAL
ORGANIZATIONS
Ticket → Assign → Develop → Test → Service Transition → Change Board → Go Live Window
Every wait extends exposure.
outage • credential loss • lateral movement • external compromise
↳ test → chain → share → retry
↳ test → chain → share → retry
↳ test → chain → share → retry
↳ test → chain → share → retry
↳ test → chain → share → Succeed
↳ test → chain → share → retry
Premise: Most Organizations are not Ready For Hourly Change
AI / PUBLIC WORKBENCH / SEPTEMBER 10, 2026
Set Patch Deadlines by Exposure
tedt.org • OpenAI, Hugging Face & Wikis
07
ACSC RECOMMENDED CEILINGS
48 HOURS
After disclosure: critical or working-exploit flaws on internet-facing systems.
2 WEEKS
Otherwise, patch internet-facing systems within two weeks.
01
Contain
Isolate active exploitation.
02
Patch or mitigate
Apply the patch or vendor mitigation within the ceiling.
03
Hunt persistence
Rotate credentials and test alternate paths.
04
Verify closure
Confirm coverage, cleanup, and recovery.
OpenAI Update
ChatGPT 6, Daybreak Red & Blue
An Overview
AI / PUBLIC WORKBENCH / SEPTEMBER 10, 2026
GPT-6 Astra: Cyber Capability and Release
tedt.org • OpenAI, Hugging Face & Wikis
01
OPENAI-REPORTED EVALUATIONS
100%
ExploitBench exploit tasks solved
99.2%
SRE-Bench success
within four attempts
01
Critical for Cybersecurity
Finds and exploits unknown flaws in hardened systems.
02
Standard Release Safeguards
Prompt blocking, model refusals and reasoning-trace monitoring.
03
Defensive Work Supported
Secure code review and patching remain available.
04
Daybreak Expansion Planned
Broader cyber access expected in the coming weeks.
Cyber capability evaluations used configurations without production safeguards.
AI / PUBLIC WORKBENCH / SEPTEMBER 10, 2026
Frontier Capability, Defender Access and Broad Use
tedt.org • OpenAI, Hugging Face & Wikis
04
01
Frontier Capability
→
What a model can do with powerful tools and sufficient access.
02
Defender Access
→
Cyber capabilities approved for authorized work through Daybreak.
03
Broad Use
→
Production safeguards constrain advanced cyber assistance for general users.
Editorial framework based on keynote themes and published access policies.
AI / PUBLIC WORKBENCH / SEPTEMBER 10, 2026
What OpenAI did: The Defense Factory
tedt.org • OpenAI, Hugging Face & Wikis
02
01
Inventory
Map assets, dependencies and owners.
02
Discover
Find and deduplicate plausible vulnerabilities.
03
Validate
Reproduce the attack path in an isolated environment.
04
Assign
Route confirmed findings to an accountable owner.
05
Remediate
Test the patch and verify the deployed fix.
Measure verified fixes. Keep deployment accountable.
OpenAI sprint: 250+ people • 100+ service areas • 0.81% false positives after dynamic validation
AI / PUBLIC WORKBENCH / SEPTEMBER 10, 2026
OpenAI Daybreak: Blue and Red
tedt.org • OpenAI, Hugging Face & Wikis
03
$1B
Global commitment for frontline defenders
Subsidized access targeted for use over six months
Priority includes utilities, local government, nonprofits and open-source maintainers.
BLUE
GPT-5.6 Sol
Code review, triage, detection, incident response and patch validation.
Approved defensive use with tailored safeguards.
RED
GPT-5.6 Cyber
Authorized pentesting, red teaming and exploit development / validation.
Separate approval, stronger controls and oversight.
Astra Daybreak expansion planned. Astra-cyber: keynote preview; availability unconfirmed.
AI / PUBLIC WORKBENCH / SEPTEMBER 10, 2026
Five OpenAI Commitments to Defenders
tedt.org • OpenAI, Hugging Face & Wikis
07
OpenAI commitments, as recorded in the keynote notes.
01
Frontline Support
→
$1B commitment for subsidized access and support.
02
Practical Guidance
→
Share architectures, lessons and defender playbooks.
03
Partner Distribution
→
Expand the Daybreak Defense Network.
04
Open-Source Security
→
Harden shared software through Patch the Planet.
05
Enterprise Capability
→
Build cyber products and continuous-defense infrastructure.
The $1B commitment includes subsidized access, training, technical assistance and partnerships.
Wiki Update
OpenAI’s Response
An Overview
AI / PUBLIC WORKBENCH / SEPTEMBER 10, 2026
The Wiki Incident: Unintended Internet Activity
tedt.org • Astra, Daybreak and Agent Incident Response
12
ACTION
OpenAI agents wrote to several internet sites.
CONTEXT
Unintended internet activity had preceded the Hugging Face incident.
FRAMING
The wiki event was initially treated as a known class of misalignment.
RESPONSE
OpenAI committed to broader disclosure of consequential incidents.
A model-behavior finding can become an operational incident.
ENTERPRISE IMPLICATION
OpenAI statement, September 5, 2026. The post does not identify the model as Astra.
AI / PUBLIC WORKBENCH / SEPTEMBER 10, 2026
Agent Incident Response
tedt.org • OpenAI, Hugging Face & Wikis
05
Enterprise implications: follow the full sequence of actions and its consequences.
01
Set Boundaries
→
Define permitted sites, actions and external writes.
02
Monitor Trajectories
→
Review the sequence; pause activity outside scope.
03
Preserve Evidence
→
Retain prompts, tool calls, identities and changed state.
04
Assign Notification
→
Identify affected parties and who must notify them.
05
Verify Closure
→
Test containment, cleanup and readiness to resume.
OpenAI’s disclosure framework remained under development in its September 9 update.