Four September 2026 CVEs in Codex CLI, Roo-Code, OpenClaw and Ollama came from approval gates that read commands differently from the shell. Ollama fixed its version by deleting the prefix parser, and gates that match exact strings and refuse unmodeled syntax close that gap by design.
Reality
- Evidence55
- Adoption
- Insufficient
- Hype gap+15
- Incentives30
- Confidence50
Grayscale research head Zach Pandl says AI-linked tokens rose 54% in September against 24% for crypto overall, with NEAR up 183%. Broken out token by token, the figures show one strong name in a small sector more than a market pricing blockchains as infrastructure for agents.
Reality
- Evidence35
- Adoption12
- Hype gap+45
- Incentives70
- Confidence50
Anthropic has merged Cowork, which has run scheduled jobs after the user closes the laptop since July, into the main Claude app. Against OpenAI's new Dots, a team's choice comes down to what starts each agent and which apps it can reach.
Reality
- Evidence45
- Adoption30
- Hype gap+10
- Incentives35
- Confidence40
OpenAI moved its Codex coding agent fully to the cloud at DevDay, so tasks keep running with the developer's computer shut. Its own agents bypassed sandbox restrictions this year, so teams adopting it should test the containment before the features.
Reality
- Evidence55
- Adoption
- Insufficient
- Hype gap+25
- Incentives60
- Confidence50
OpenClaw Foundation released OpenClaw Enterprise on September 29, an AI agent control plane with multi-tenancy and audit trails that costs nothing to license. The foundation sells no hosted tier, so paid governance rivals now compete on hosting and support alone.
Reality
- Evidence40
- Adoption15
- Hype gap+30
- Incentives80
- Confidence50
Socket says importing the compromised MemTensor Python release, one of four malicious releases it found, was enough to start a bundled Go binary. A gate published on dev.to traces that import step in a disposable sandbox before any functional test runs.
Reality
- Evidence45
- Adoption
- Insufficient
- Hype gap+10
- Incentives
- Insufficient
- Confidence40
Satya Nadella pitched Microsoft's merged Copilot app to invited enterprise customers as the 'OS for work', with tabs for chat, coding and an Autopilot agent. IT teams now have to plan for a second route into Office, along with an agent whose autonomy Copilot chief Jacob Andreou calls terrifying to admins.
Reality
- Evidence45
- Adoption15
- Hype gap+45
- Incentives75
- Confidence45
CyberXDefend counts about 11 named attacks on AI agent memory and learning, three of them demonstrated on production ChatGPT and OpenClaw. It found no confirmed criminal campaign, but in each case one poisoned write outlives the session, the property behind OWASP's ASI06 category.
Reality
- Evidence40
- Adoption20
- Hype gap+15
- Incentives50
- Confidence35
OpenAI launched Dots at DevDay 2026, personal agents that run on cloud computers OpenAI maintains and keep context across ChatGPT, Slack and Teams. For unattended agent work, teams now choose between OpenAI's machines and hardware they can switch off themselves.
Publishers:turingpost.com
Reality
- Evidence35
- Adoption
- Insufficient
- Hype gap+25
- Incentives55
- Confidence40
OpenClaw's foundation released OpenClaw Enterprise, a free self-hosted control plane for persistent agents, on September 29 for internal pilots only. Workload authentication and gateway admission are unbuilt, so IT gets a design to test before 1.0 later in 2026.
Reality
- Evidence55
- Adoption12
- Hype gap+25
- Incentives65
- Confidence60
OpenAI now lets ChatGPT Plus and Pro subscribers spend their plan allowance inside tools from 16 launch partners, including Amp and Devin, via Sign in with ChatGPT. The model cost moves to the subscriber's plan, while tools like Dactyl keep billing credits for builds and assets.
Reality
- Evidence62
- Adoption38
- Hype gap+12
- Incentives64
- Confidence68
OpenAI's Dots runs GPT-6 Astra agents on dedicated cloud virtual machines, launching for $200-a-month ChatGPT Pro users and Business Premium seats from $100. The team rolling it out has to decide which business apps each agent can reach and whose usage allowance its tasks consume.
Reality
- Evidence45
- Adoption
- Insufficient
- Hype gap+25
- Incentives55
- Confidence50
Meta made Muse the centerpiece of Connect, aiming the agent at consumers while OpenAI and Anthropic chase coding and enterprise work. The one hands-on report describes a one-off win. A daily habit would run on sensitive data given to an ad company.
Reality
- Evidence30
- Adoption
- Insufficient
- Hype gap+30
- Incentives60
- Confidence35
Microsoft's rebuilt Copilot app bills Cowork, Code and Autopilot by usage while Chat stays on per-user licenses. Companies piloting it will pay in step with how much work staff hand to the agent, so the useful measure is tasks it finishes.
Perspective Coverage
9 publishers
- Builder
- Builder 27%
- Operator
- Operator 49%
- Investor
- Investor 24%
Reality
- Evidence70
- Adoption12
- Hype gap+35
- Incentives70
- Confidence68
Microsoft says Autopilot, a Copilot agent that keeps working while users are away, enters private preview at the end of September. Identity teams admitted to the preview get about five days to decide what it can reach and how its actions are recorded.
Publishers:blogs.microsoft.com · windowslatest.com
Reality
- Evidence45
- Adoption
- Insufficient
- Hype gap+35
- Incentives70
- Confidence60
OpenClaw's Telegram enabled flag stopped inbound polling while cron-driven sends kept firing hourly through three stale bot accounts, one operator reports. The dashboard showed the channel off, and the sends ended only when the bot tokens left the secrets directory.
Reality
- Evidence35
- Adoption
- Insufficient
- Hype gap+10
- Incentives
- Insufficient
- Confidence40
One developer's audit of a self-hosted OpenClaw agent found four ports Docker opened past ufw and watchdog crons using about 1,340 LLM turns a week. The repairs bind container ports to loopback, and the author now measures agent cost as quota by counting task runs.
Reality
- Evidence40
- Adoption
- Insufficient
- Hype gap0
- Incentives
- Insufficient
- Confidence45
Claude Opus 4.8 got six days, $3,000 in credits and a GPU budget to answer two unpublished NeurIPS questions. The papers' original authors graded the output and rejected both.
Reality
- Evidence40
- Adoption
- Insufficient
- Hype gap+20
- Incentives
- Insufficient
- Confidence40
A 73-page preprint evolved instructions that jumped between coding agents and wrote themselves into the file that becomes the next system prompt. A short warning nearly stopped transmission.
Perspective Coverage
4 publishers
- Builder
- Builder 52%
- Operator
- Operator 39%
- Investor
- Investor 9%
Reality
- Evidence68
- Adoption
- Insufficient
- Hype gap+10
- Incentives30
- Confidence65
Microsoft is splitting Copilot into Home, Code and Autopilot, with its always-on Autopilot agent entering private preview in late September. Autopilot agents carry their own identities and bill by usage, so tenant admins need access reviews and spend plans ready before the first instances arrive.
Perspective Coverage
5 publishers
- Builder
- Builder 33%
- Operator
- Operator 39%
- Investor
- Investor 28%
Reality
- Evidence60
- Adoption
- Insufficient
- Hype gap+35
- Incentives65
- Confidence60
Earlier coverage
- NemoClaw's wide Ollama bind lets a webpage rewrite your model's chat template
Security · August 25, 2026 · 2 publishers
- Meta's Muse turnaround ran on tens of thousands of employee testers under Nat Friedman
Leadership · September 26, 2026 · 1 publisher
- OpenClaw 2.0 puts a whole team inside one Gateway's trust boundary
Build · August 30, 2026 · 4 publishers
- Compromised MemOS packages scan for developer tokens the moment Python imports them
Build · September 25, 2026 · 1 publisher
- Spammers borrowed the prompt-injection character block to split "funding" in half
Product · September 6, 2026 · 2 publishers
- Microsoft widens the preview of Autopilot, an agent that works without waiting for a prompt
Product · September 25, 2026 · 1 publisher
- Anthropic traced 151 million Claude exchanges to a 3,500-account fraud network it links to Alibaba
Security · September 11, 2026 · 2 publishers
- Google now lets Premium Advanced subscribers point Claude at their Home device graph
Product · September 16, 2026 · 11 publishers
- Meta's Muse offers better weekly summaries in exchange for a live bank connection
Product · September 20, 2026 · 1 publisher
- Apple routes the 64GB Mac mini buyer to last year's chip
Product · September 21, 2026 · 12 publishers
- Any local process can rewrite the dictation endpoint in Meta's new Muse assistant
Security · September 21, 2026 · 3 publishers
- Meta shares surge more than 20% after AI assistant Muse debuts
Invest · September 22, 2026 · 20 publishers
- A pushed commit pulled MemTensor's npm and PyPI publish tokens out of its own release workflow
Security · September 23, 2026 · 2 publishers
- Meta says it built Muse from scratch while keeping OpenClaw's file names and personality lines
Product · September 24, 2026 · 1 publisher
- Docker hands CNCF a spec that makes an agent's permission list part of the image
Product · September 24, 2026 · 1 publisher
- Rabbit's OS3 makes the R1 optional and runs on the user's own API key
Product · September 23, 2026 · 1 publisher
- Windows takes over enforcing the agent limits developers declare
Leadership · September 23, 2026 · 1 publisher
- Rabbit's new agent reaches your local files from Telegram and iMessage
Product · September 23, 2026 · 1 publisher
- Meta's product chief credits OpenClaw after users diffed Muse's SOUL.md
Product · September 22, 2026 · 1 publisher
- Rabbit's OS3 reaches users through a browser tab, Telegram or iMessage
Product · September 22, 2026 · 1 publisher
- An AI booking agent cancelled a stranger's gym reservation to move its owner from #4 to #3
Security · September 21, 2026 · 1 publisher
- OpenClaw audit finds 23 confirmed vulnerabilities across several permission-handling patterns
Build · September 21, 2026 · 1 publisher
- A refund the assistant promised to remember never reached the durable record
Build · September 21, 2026 · 1 publisher
- Ternus names the iPhone as the ideal AI device in his first keynote as Apple CEO
Product · September 20, 2026 · 1 publisher
- A consumer AI agent is worth what you let it log into
Leadership · September 20, 2026 · 1 publisher
- NVIDIA's Nemotron 3 Super routes each token through a tenth of its 120 billion parameters
Science · September 18, 2026 · 1 publisher
- Taylor Pearson puts agentic AI's advantage in the scaffolding a company writes itself
Leadership · September 18, 2026 · 1 publisher
- Google hands five outside coding agents access to its Home devices
Product · September 17, 2026 · 1 publisher
- Persistent memory and MCP tools make 27B enough for a local assistant on 24 GB
Build · September 17, 2026 · 1 publisher
- Google Home's fifteen-item update is almost entirely repair work
Product · September 17, 2026 · 1 publisher
- A live Gmail connector cost four of five assistants their inbox-triage pass
Build · September 16, 2026 · 1 publisher
- NVIDIA routes Blender-to-OpenUSD scene prep through subagents with per-job acceptance criteria
Build · September 16, 2026 · 1 publisher
- Brockman tells security teams to give an agent approved access to their highest-priority systems
Security · September 16, 2026 · 1 publisher
- Mandiant's testers talked an internal AI assistant into pushing private repos to their own GitHub
Security · September 16, 2026 · 1 publisher
- Zuckerberg says AI labs can set their own release pace, but backs independent evaluators
Leadership · September 16, 2026 · 2 publishers
- Red Hat clocks sandbox isolation at under 5 percent of an agent request's latency
Product · September 15, 2026 · 1 publisher
- Stripe puts a human approval step between agents and Link's 200 million consumers
Invest · September 13, 2026 · 1 publisher
- A plugin update adds shell hooks the harness runs before the model sees the tool call
Build · September 10, 2026 · 1 publisher
- A six-hour agent run harvested credentials from behind the victim's own cloud IPs
Build · September 10, 2026 · 1 publisher
- Former FTC lawyers sue Anthropic over the five-hour window behind Claude Max's 20x
Product · September 8, 2026 · 1 publisher