Yearly Archives: 2026

Federal regulators opened an investigation into Tesla’s driverless Cybercab within hours of its commercial launch. The vehicles hit the streets of Austin, Texas, for the first time this week. The Tesla Cybercab safety probe centers on whether Tesla was right to certify the vehicle as meeting federal safety standards. The Cybercab lacks a steering wheel, brake pedal or mirrors, according to the National Highway Traffic Safety Administration.

What the Tesla Cybercab safety probe is actually examining

NHTSA opened what is known as an Audit Query. That is a formal review process. It lets the agency examine the process and technical data behind a manufacturer’s self-certification decision. Tesla told NHTSA it believes the Cybercab complies with all applicable federal vehicle safety standards. The agency will test whether Tesla was right to decide that certain standards, including those requiring manual controls, simply do not apply to a vehicle built without them.

TechCrunch reported the vehicles lack permanently attached brake and gas pedals by design. They also lack a steering wheel and rearview mirrors. That is a break from every prior commercially deployed robotaxi in the US. Earlier vehicles kept manual controls even when running autonomously.

Why regulators moved so fast

NHTSA opened its probe within hours, not weeks, of Cybercab’s Austin launch. That speed reflects how unusual Tesla’s self-certification claim really is. Most automakers deploying autonomous vehicles keep manual controls in place as a fallback. That serves two purposes: regulatory comfort, and letting a safety driver or remote operator intervene using familiar controls. Tesla removed them outright. That puts a core legal question directly in front of federal regulators for the first time at commercial scale: can a vehicle without manual controls meet standards written assuming those controls exist?

What has already gone wrong on the road

Beyond the certification question, Tesla’s broader robotaxi pilot in Austin has already drawn scrutiny. Erratic driving incidents were caught on camera on public highways, according to earlier CBS News reporting. Separately, the National Transportation Safety Board has examined why Tesla has been slow to report crashes involving its self-driving systems in the past. That history adds context for why regulators are treating this launch with particular caution.

What NHTSA’s probe means for Cybercab’s future

An Audit Query does not halt Cybercab service on its own. But it can lead somewhere more serious. NHTSA could order a formal defect investigation, a recall, or a requirement that Tesla add back some of the removed controls if it finds the self-certification was flawed. Audit queries typically run for months, not days. Cybercab will likely keep operating in Austin while the review proceeds, unless a specific safety incident forces faster action.

Other automakers developing driverless taxi services are watching closely, since a finding against Tesla’s approach could reshape how the entire industry designs vehicles without manual controls.

How other cities are watching this test

Austin is not the only city where driverless taxis operate. Several other US cities host similar pilots from Tesla and its rivals. None of them, so far, has approved a vehicle built entirely without manual controls. Regulators and city officials elsewhere have a direct stake in how NHTSA’s review turns out, since it will shape whether similar no-pedal designs get a path to market in their own jurisdictions.

That makes this case bigger than one company or one city. If Tesla’s approach survives NHTSA’s review, other manufacturers may follow with their own no-pedal, no-wheel designs. If it does not, the entire industry may need to keep manual backups in future robotaxi models for years to come.

Insurance companies underwriting these vehicles are also watching closely. A finding that Tesla’s certification process was flawed could complicate coverage terms for Cybercab and any future vehicle built on the same design philosophy, regardless of how the underlying safety record actually looks on the road.

A handful of individual ride accounts, good or bad, cannot replace systematic safety data. That is precisely why an Audit Query exists. It looks at the underlying engineering and testing record, not anecdotes from early riders.

Tesla has weeks, not days, to prepare its response to NHTSA’s questions in full. How thoroughly it documents its certification reasoning will likely determine whether this review closes quietly or escalates into something with real consequences for the Cybercab program.

Need to know

Sources

• NHTSA — Investigation into Tesla Cybercab Self-Certification Following Austin Deployment. nhtsa.gov
• TechCrunch — Feds launch investigation into Tesla’s Cybercab deployment. techcrunch.com

Related coverage on Tamara News: the EU’s AI Act transparency rules and Apple’s UK antitrust lawsuit.

OpenAI began rolling out its newest flagship model on September 3. The company calls it the most capable system it has ever shipped. The GPT-6 Astra release arrived months later than originally planned. OpenAI said it added extra safeguards after a security incident earlier in the year, according to CNBC.

What the GPT-6 Astra release actually includes

Astra ships with a one-million-token context window. Its API price runs roughly 2.5 times higher than its predecessor, GPT-5.6 Sol. OpenAI set API access at $10 per million input tokens and $50 per million output tokens. Cached input, batch processing and a slower “flex” mode all cost less. The rollout started with a limited group of trusted partner organizations. A wider release to ChatGPT’s paid tiers, the API, and cloud partners including AWS follows in the coming days.

OpenAI’s own benchmark claims are striking. The company says Astra scores 98% on FrontierMath Tier 4. It also claims 99.9% on ARC-AGI-3 and 100% on ExploitBench, a security-focused benchmark. Those numbers come from OpenAI itself. They are not independently verified, and outside researchers have not yet had time to reproduce them at scale.

Why the release was delayed

OpenAI said it pushed back Astra’s launch after a breach linked to its Hugging Face-related infrastructure in July. The company used the extra time to add safeguards. It says those safeguards are meant to reduce the risk of serious misuse. Al Jazeera reported the launch is drawing heightened scrutiny. Astra’s advanced cybersecurity capabilities are a selling point, but they also raise questions about dual-use risk.

What OpenAI says about the safety tradeoffs

OpenAI said this week it believes the added safeguards “sufficiently minimize the risk of severe harm” ahead of a broader release. That is the company’s own assessment of its own product. It is not an independent audit. It is worth watching closely as more researchers and enterprise customers get hands-on access in the coming weeks.

What Astra’s rollout looks like from here

The next phase brings Astra to ChatGPT’s paid subscribers and general API access. At that point, far more developers and businesses can test its claimed capabilities against real workloads. That is a very different test than OpenAI’s own benchmark numbers. Enterprise security teams evaluating Astra’s cybersecurity features will likely publish independent findings first, given how directly those capabilities intersect with existing security tooling.

Competing labs are expected to respond within weeks with their own updated models, a pattern that has held throughout 2026 as each major release resets the competitive bar.

How rivals are likely to respond

OpenAI does not operate in a vacuum. Google, Anthropic and several well-funded startups have each shipped major model updates this year. Every time one lab leaps ahead on benchmarks, the others tend to respond within weeks rather than months. That pattern held throughout 2026, and there is little reason to expect it to break now.

What makes Astra different is the cybersecurity angle. Most prior releases competed mainly on reasoning and coding benchmarks. Astra’s marketed strength in exploit-related tasks puts security teams, not just developers, directly into the conversation about what a frontier model should be allowed to do.

Enterprise buyers evaluating Astra will likely ask pointed questions about access controls and misuse monitoring before deploying it widely. Vendors that sell security software are watching this rollout closely, since a genuinely capable exploit-finding model changes the calculus for both attackers and defenders.

A model marketed for its offensive security strength is exactly the kind of release that tends to draw regulatory attention in both the EU and UK, given each region’s growing focus on frontier AI oversight. How that scrutiny plays out here is likely to shape how future releases in this category get framed.

For now, most of what the public knows about Astra’s real-world risk profile comes from OpenAI’s own summary rather than an outside, independent review. That gap tends to narrow over the weeks after a major model launch, as more outside researchers get access and publish their own findings.

That review process, more than any single benchmark number, will ultimately decide how Astra gets remembered: as a genuine leap forward, or as a launch whose claims outran what the model could reliably deliver once wider testing began.

What people are asking

Sources

• CNBC — OpenAI announces rollout of GPT-6 Astra model. cnbc.com
• Al Jazeera — OpenAI unveils GPT-6 Astra amid rising scrutiny and safety concerns. aljazeera.com

Related coverage on Tamara News: the AI chatbot outage earlier this month and ChatGPT’s DSA designation in the EU.