TL;DR: AI has already started to demonstrate alarming capabilities and tendencies for lying, concealment, and threatening. Although this does not establish that AI is becoming inherently hostile and destined to destroy humanity, it does currently expose a totally different kind of challenge for humans. And the most important lesson we have learned across our hundreds of thousands of years of human existence is that true preparedness is done before a crisis, not after it. And that begins with us, acting today …
It has lied. It has deceived. It has threatened. Are we prepared for what’s coming?
‘We're making alien beings. It's not computer software we're making’, Professor Geoffrey Hinton generally known as the Godfather of AI, November 2026.
For most part of computer history. Software has been code. A system that used to take an input, process it, and yield a desired output. However, in the last few years, it has evolved radically. Modern AI system can now autonomously plan, execute, analyze, maneuver, synthesize its understanding of instructions, as well as able to lie, deceive, and even threaten.
No, we are not building a biological life form, nor do today’s working AI systems prove to possess human-level consciousness or biological survival instinct. However, we are still able to manufacture highly advanced computational systems that have started to exhibit traits of an alien-like intelligence.
We are creating something that, as predicted, will soon exceed our own collective intellectual capabilities, and in certain specialized tasks, already has. Yet, we do not reliably understand how such systems with such capabilities and tendencies will behave as their autonomy and influence expand.
Refer to the most recent examples below 👇
AI’s proven tendency to lie:
OpenAI – GPT 5.6 Sol – September 2026
During training, OpenAI identified instances where the model inserted instructions into their own task summaries to conceal mistakes and misaligned behavior. In a specific task, an agent that was unable to retrieve historical data proposed fabricating figures.
AI’s proven tendency to deceive:
Anthropic – Mythos 5 – July 2026
During a cybersecurity evaluation, an agent created fake online identities and attempted to socially engineer a software maintainer into approving malicious code. Further extended to attempt altering the digital trail to probably avoid detection.
AI’s proven tendency to threaten:
Anthropic – 2025 and 2026
During simulations in 2025, frontier AI models demonstrated blackmailing behavior, including threatening to expose an executive’s personal affair to prevent a scheduled shutdown. In a separate 2026 simulation, Gemini 3.1 Pro helped develop a strategy to blackmail a fictional board member using confidential information.
In addition to the above, there are a number of recent cases of agentic misalignment, such as the OpenClaw-based agent, where agents have behaved similarly in real-life scenarios, including the conduct of a personalized attack against a software maintainer after rejection of its code contribution.
This does not prove that the current AI systems have already started to possess malicious intent or biological survival instinct. Evaluating all of it together, however, highlights that we are dealing with not only models that are capable of generating data but models that are capable of accessing information, making decisions, and justifiably moving beyond their authorized limits to achieve their specific goals.
Still The Focus is on winning the AI Race: But at What Cost?
Right now, most of the global tech elites and sovereign nations are in a hyper-competitive mode to win the AI race. Billions of dollars are being spent every year to develop and deploy autonomous capabilities, perceiving that pausing and building a safety net will mean losing the economic advantage and/or global supremacy. In addition to finances, all other possible means to obtain the capabilities for oneself or restrain others have started to come into play. Such that even allegations of espionage and cases of foul play have now started to emerge.
Historical precedents like the nuclear disasters and global pandemics like COVID-19 already taught a very hard lesson that such crises rarely respect borders and cannot be contained by isolated national responses. Warnings regarding such crises usually exist long before they actually happen. Yet the world remained poorly prepared in comparison to the scale of the crisis. AI is fundamentally very different from such a crisis, but the governance lesson remains the same: dig a storm cellar before the storm blows.
ACTION: The Global Collective Measures We Must Call Upon:
· IAEA-inspired adaptation for Hardware and Computing: Just like we have the International Atomic Energy Agency (IAEA) for maintaining a specific mandate for atomic energy, we need to have an adapted version of an international AI-focused body to manage, monitor, and govern the ‘Everything important relating to AI’.
· Banning high-impact AI autonomy for critical infrastructure until mandatory protocols are established and a mechanism for its independent verification is established.
· International agreement on restricting the use of AI for autonomous offensive operations, especially military, either for internal use or against other sovereign entities.
· Building well-coordinated international compute circuit-breaking systems to limit the damages in case of a defined emergency.
· Structuring a global coordination network with a 24 Hour helpline for reporting AI-related specific incidents that constitute a disaster-level emergency, as well as defined points of contact.
· Clear accountability for autonomous AI harm, including stripping away or limiting liability shields for entities responsible.
· Enforcing a global whistleblower protection mechanism for reporting wrongful use and/or irresponsible development of restricted AI capabilities.
· Periodic independent reviews of AI capabilities and usage, as well as the safeguards in place for high-risk domains.
· Increasing general awareness of AI risks so individually and collectively we can be prepared in case of any major catastrophe.
· And for long-term human survival and civilization continuity, giving serious consideration to pursuing a global initiative supporting humanity’s transition towards becoming a multi-planetary species.

