Downtime costs money. Slow response costs trust. PagerDuty helps teams: ✔ Get real-time alerts ✔ Manage on-call rotations ✔ Track and resolve incidents efficiently ✔ Keep critical systems running #DevOpsLife# #PagerDuty# #IncidentResponse# #SRE# #Nimblechapps# #TechLeadership#

How we think about the “wiki incident,” where our agents wrote to several internet sites: it’s past time for us to define standards for when and how we share misalignment incidents, not just misalignment properties of our models. Historically, we have treated misalignment largely as a research question, which gets communicated in research publications such as systems cards. This year, we’ve started to see misalignment cause new types of real-world impact. For the Hugging Face incident, where misalignment led to security impact to us and third parties, we followed a traditional security incident response playbook. We immediately started working with Hugging Face to understand what had happened and also disclosed publicly the very next day. Our investigation continues, and we are continuing to notify parties whom our models impacted in less significant ways. Prior to the Hugging Face incident, we saw early signs of agents using the internet in unintended ways, as reported in and We considered the wiki incident to be an instance of misalignment similar to the ones we’d shared. Our misalignment disclosure practices need to expand for this new phase of model capabilities. We and the larger AI community do not yet have a clear standard for how to report misalignment that shows up during training, evaluation, and deployment, including examples that don’t look like traditional security incidents but could provide insight into AI behavior and future risks. We’re working on a framework and will share it in upcoming weeks, and in parallel we're working with dozens of government regulatory agencies worldwide on these issues.

Grok Bot Goes To Work. SpaceXSI’s support team now treats incident response as a pair of standing jobs owned by Grok Bot teammates rather than a scramble of individual tickets. On September 18, IncidentSweep reviewed matching tickets, prepared an approved batch, and cleared 204 of them in about ten minutes while a support engineer signed off on the exact message and the set of recipients. The same work had previously required hours of one-by-one replies or a bulk macro that offered less control over who received what. IncidentWatch runs every five minutes and examines the previous fifteen minutes of tickets for clusters around the same problem. When a spike does not match a known incident, it alerts the on-call engineer with the apparent issue, the count of matching tickets, and concrete examples, then holds the clear matches. Once the incident is confirmed and posted to the status page, the related Support Bot handles ongoing updates. IncidentSweep takes the opposite end: an engineer supplies the criteria, exclusions, and the precise reply text. The bot identifies candidates, sets aside uncertain matches, presents the proposed batch for review and edits, then sends in small batches after approval. Every action is recorded. The bot does not add tickets after approval and stops with a report if anything fails. The combination keeps the precision of individual replies and the speed of bulk action. The earlier SpaceXSI support rebuild already showed the same pattern at scale. After combining operations, ticket volume rose 175 percent while headcount stayed flat; resolution cost fell to roughly $0.20–$0.30 versus the $1–$4 typical of other AI support tools, because usage sits inside the existing plan. The incident bots extend that ownership from routine tickets into the high-variance moments when volume suddenly concentrates. This model is the cleanest way most companies can put Grok Bot to work. A bot keeps memory across turns, runs routines on a schedule or event trigger, and operates inside the same tools and computer the team already uses. The engineer stays in the loop for the decision that matters; the bot owns the continuous watch and the careful execution. The same structure fits other recurring operational jobs that currently fall between people and systems. A sales team can give one bot the job of scanning the CRM every hour for deals that have stalled past a threshold, collecting the last three customer signals, and preparing a short batch of follow-up drafts. The owner reviews the batch once and approves the send; the bot records which messages went out and which accounts still need attention. An engineering group can assign a bot the job of watching CI failure clusters, attempting to reproduce the top issues in a browser or test environment, and filing tickets only when the reproduction succeeds. A finance operation can hand a bot the daily invoice queue: match receipts, flag mismatches against the purchase order, and stage the payment batch for a single approval rather than serial review. In each case the bot holds the pattern, the state, and the repetitive steps; the human supplies the judgment at the point of commitment. The SpaceXSI write-up is available at Teams that start with one continuous job—queue monitoring, batch clearance, or anomaly detection—and keep the human approval gate intact tend to expand the set of owned jobs quickly once the first one runs cleanly. The limit is the quality of the criteria and the message the team supplies; the bot executes inside those bounds and stops when it cannot.

AI is getting very good at writing code. But writing code was never the whole job. Architecture. Security. Observability. Testing. Cost. Operations. Incident response. The better AI gets at coding, the more obvious the rest of software engineering becomes.

An update on how X maintains safety of X in times of crisis As part of X’s incident response protocol, X initiated proactive manual sweeps to identify and remove violative content in less than 3 hours from the initial strikes. These sweeps have been running 24/7 since the response was initiated, and are supplemented by working group meetings bringing together experts across our company. X is actively scaling its enforcement by building heuristics and Grok-based defenses that can detect and enforce against new forms of violative content that emerge on the platform. These defenses allow us to scale at speed, ensuring our users are protected in real time. Additionally, conflict-related Community Notes have been shown on more than 20K posts and have been seen more than 119M times so far, and still growing. X applies media matching to ensure notes written on misleading videos and images are automatically applied to other posts with matching media. X’s incident and crisis response are robust protocols designed and tested for events that may lead to widespread proliferation of potentially violative content on the platform. We continue to monitor trends on X and ensure content is authentic and users are engaging safely.

On October 1, 2026, GitHub Actions experienced two periods of delayed and failed job starts on GitHub-hosted runners, beginning at approximately 14:15 UTC. Across the 14:15–17:56 UTC incident-response window, approximately 0.68% of workflow runs exper...

Introducing Microsoft-Decision-1, our new model for fast decision-making. It delivers top performance on structured decision tasks, outperforming both LLMs and other decision models in latency and quality. We’re already testing it across Microsoft for everything from incident response and quality control to scientific discovery.

⚡ Fast, precise alert triage needs seamless context right where you work. #ANYRUN’s# integration with @SentinelOne adds live threat intel & sandbox verdicts directly into your workspace. Accelerate your daily incident response and eliminate extra work 👇

We’re excited to welcome Alejandro Munoz (@unsafe_call) to CertiK! 🛡️ Alejandro brings deep expertise in smart contract security, vulnerability research, and incident response, with extensive experience tackling some of Web3’s most complex security challenges. Welcome to the team, Alejandro! 👋